Re: [PATCH 4/9] x86, NMI: Remove DIE_NMI_IPI and add priorties to handlers

[Cyrill Gorcunov <gorcunov@gmail.com>]

On Tue, Nov 30, 2010 at 05:27:25PM -0500, Don Zickus wrote:
> When re-ordering how the NMI handles its callbacks, a conversation started
> asking what DIE_NMI_IPI meant.  No one could answer it.

It should have came from commit

 | commit c4b2bffee2a4115fed2825530f2b906ee2f17bd7
 | Author: Andi Kleen <ak@suse.de>
 | Date:   Fri Jan 23 18:46:40 2004 -0800
 |
 |   [PATCH] x86-64 merge
 |   
 |   Mainly lots of bug fixes and a few minor features. One change is that
 |   it uses drivers/Kconfig now like i386. This requires a few minor changes in
 |   outside Kconfig files which I am sending separately.
 ...

Andi do you remember what the initial idea was? Didn't find any user of it
even in this old commit. Just curious.

> 
> Noticing that is was wasteful to call the die_chain a second time with just
> another argument, DIE_NMI_IPI, it was decided to nuke it and add priorities
> to the die_chain handlers to maintain existing behaviour.
> 
> This patch replaces DIE_NMI_IPI with the appropriate option, mostly DIE_NMI.
> Then it adds priorities to those handlers, using a globally defined set of
> priorities for NMI.
> 
> The thought is eventually we will just switch the nmi handlers from the
> die_chain to something more nmi specific.
> 
> Signed-off-by: Don Zickus <dzickus@redhat.com>
> ---

 Don, maybe switching to say new chains like chain_perf and friends would be
more readable/clean? I'm not against this patch by any means, but just a thought ;)

 Ie I thought like

 default_do_nmi
   if (!(reason & 0xc0)) {
     if (notify_perf() == NOTIFY_STOP)
       return
     if (notify_die() == NOTIFY_STOP)
       return
    ...
   }

 Or there is something obvious I'm missing?

 Again, just a thought.

 Cyrill