RE: [v2 PATCH 1/3] panic: Change nmi_panic from macro to function

[æåèå / KAWAIïHIDEHIRO]

Hi,

> From: Borislav Petkov [mailto:bp@xxxxxxxxx]
> On Wed, Mar 02, 2016 at 02:18:24PM +0100, Michal Hocko wrote:
> > On Wed 02-03-16 19:36:26, Hidehiro Kawai wrote:
> > [...]
> > > +void nmi_panic(struct pt_regs *regs, const char *fmt, ...)
> >
> > Do we really need vargs? All the current users seem to be OK with a
> > simple string. This makes the code slightly more complicated without any
> > apparent reason.
> 
> I was just wondering the exactly same thing...
> 
> The contra-arg would be that in case someone wants to do nmi_panic()
> with more than a string, then it won't work.

It's not necessary to use vargs at this point, and passing a simple
string is OK for me.  Even if someone wants to use vargs, we can modify
nmi_panic() without any changes in caller side.  So, I'll remove it.

> The question is, does nmi_panic() even need to dump something more than
> regs and a string?

Hmm, printing regs, especially for RIP, would be useful for hang-up
cases, but I don't want to do much things in nmi_panic() as long as
it is already done by current callers.  nmi_panic() can be called
concurrently on multiple CPUs, so it also needs another serialization
to print something.

By the way, I have a patch set to safely leave important information
by kexec's purgatory...but no one is interested in?
http://thread.gmane.org/gmane.linux.kernel.kexec/15382

Regards,
--
Hidehiro Kawai
Hitachi, Ltd. Research & Development Group