Re: [RFC][PATCH -next] make pstore/kmsg_dump run after stopping other cpus in panic path

[Don Zickus <dzickus@redhat.com>]

On Thu, Oct 20, 2011 at 11:13:26AM -0400, Seiji Aguchi wrote:
> Don,
> 
> >That is part of the wider problem with kmsg_dump that Vivek talks about
> >with me, is that it is just a giant hook in the panic path with limited
> >auditing.  So we need to explicit set our expectations with BUG_ONs/WARNs
> >otherwise we might get bit later by them.
> 
> I found an issue while developing a patch v2.
> 
> We can't call BUG_ON(in_nmi() && reason != KMSG_DUMP_PANIC)
> because kmsg_dump(KMSG_DUMP_EMERG) is called in NMI context if "panic" kernel parameter is set.

I am confused.  Did you mean 'panic' kernel parameter is _not_ set?

Actually, BUG_ON is probably overkill, perhaps a WARN_ON would be more
appropriate.

> 
> I will keep doing research bust_spinlock() or vprintk() so that lockdep checking works and we can avoid any deadlocks.

I wonder if it would be smarter to split your patch in half.  The part
where you move kmsg_dump to below stop_cpus is probably non-controversial.
That might make it sooner than the more controversial bust_spinlock
pieces.

Cheers,
Don