Re: Dealing with the NMI mess

[Linus Torvalds <torvalds@linux-foundation.org>]

On Fri, Jul 24, 2015 at 1:51 PM, Peter Zijlstra <peterz@infradead.org> wrote:
>
> do_debug()
>   send_sigtrap()
>     force_sig_info()
>       spin_lock_irqsave()
>
> Now, I don't pretend to understand the condition before send_sigtrap(),
> so it _might_ be ok, but it sure as heck could do with a comment.

Ugh. As Andy said, I think that's ok, because it's actually the
single-step case, and won't trigger for kernel mode. So we should be
ok. Although the code I agree is not good.

I'd personally be more worried about the usual crazy "notify_die()"
crap. I absoluely detest those notifier chain things. They are hooks
for random crap that shouldn't be hooked into, but whatever. It's not
a problem in practice, it's just a sign of a certain kind of diseased
mind.

On the whole I think we're ok. I'd love to get rid of things, and yes,
I think we should probably explicitly handle the in-kernel case first
and just return without doing anything, just to make the code more
obviously safe. But it doesn't look like a fundamental problem spot.

                  Linus