Re: movidis x16 hard lockup using 2.6.33

[David Daney <ddaney@caviumnetworks.com>]

On 03/29/2010 03:02 PM, Andreas Barth wrote:
> * David Daney (ddaney@caviumnetworks.com) [100329 18:54]:
>> On 03/27/2010 04:07 PM, Andreas Barth wrote:
>>> * David Daney (ddaney@caviumnetworks.com) [100326 19:57]:
>>>> Also you could try running with the attached patch.  It is not the best
>>>> watchdog, but it will print the register state for each core when things
>>>> get stuck.  Occasionally that is enough to see where the problem is.
>>>
>>> Thanks.
>>>
>>> As our logging has only limited buffer size, I'd be happy about an
>>> variant of the patch which doesn't reboot but just let the machine
>>> hang after the third occurence.
>>>
>>> Any chances for it?
>
>> You could just sit in a loop kicking the watchdog timer after you get to
>> the NMI handler.  That should prevent a reset, but still print the
>> machine state.
>
> I need to admit that I'm totally unable to make code from that
> statement.
>
>
> Could you (or someone else) give me a hand? Also please note that it
> usually takes a few hours to crash the machine, and I didn't see
> anything in the normal syslog.

At the end of octeon_watchdog_nmi_stage3, instead of returning, do:

for(;;) watchdog_poke_irq(0, NULL);

That should prevent it from rebooting.  The messages will appear on the 
serial port, not in syslog.

David Daney