SW_WDOG problem

[Scott Long]

Gerrit Nagelhout wrote:

> Hi,
> 
> Were are trying out the new software watchdog on current (from a couple of
> weeks ago) and are running into a panic within a panic.  What happens is
> when the software watchdog tries to panic the system (with interrupts
> disabled), 
> it manages to write out the core file, but then panics again in
> pmap_invalidate_range
> because interrupts are enabled.
> Any ideas on how to work around this?
> Thanks,
> 
> Gerrit 
> 
> 

This is from a well known class of problems.  There has been talk about 
disabling the shutdown fsync in the panic case.  It's both unsafe
(why try to continue to do I/O when you've paniced because the
system has gone into an unreliable state) and hard to do correctly
without running into these kinds of problems.

Scott