Re: [PATCH] perf_counter: powerpc: add nmi_enter/nmi_exit calls

[Paul Mackerras <paulus@samba.org>]

Ingo Molnar writes:

> I'm wondering, what was the real impact? Was it a crash or some 
> other misbehavior? This impact line:
> 
> 	Impact: powerpc bug fix
> 
> is a bit too generic to be useful in practice. Something like:
> 
> 	Impact: fix stuck NMIs on powerpc
> 	Impact: fix NMI crash on powerpc
> 
> would have been more descriptive about the real, hands-on impact of 
> this patch.

I was looking at Peter's patches and I noticed he used in_nmi(), and I
wondered "what's that?", so I went looking and found it, and realized
that I needed to be calling nmi_enter/exit for it to work.  I never
actually booted a kernel that had the patch to use in_nmi() but not my
patch to call nmi_enter/exit.

The impact would have been that in_nmi() always returned 0, so I
expect that I would have seen deadlocks and/or memory corruption had I
booted a kernel without my fix.

Paul.