Re: [PATCH 1/5] x86, nmi: Add new nmi type 'external'

From: Ingo Molnar <mingo@kernel.org>
To: Don Zickus <dzickus@redhat.com>
Cc: x86@kernel.org, "Peter Zijlstra" <peterz@infradead.org>,
	ak@linux.intel.com, gong.chen@linux.intel.com,
	LKML <linux-kernel@vger.kernel.org>,
	"Thomas Gleixner" <tglx@linutronix.de>,
	"Frédéric Weisbecker" <fweisbec@gmail.com>,
	"Steven Rostedt" <rostedt@goodmis.org>,
	andi@firstfloor.org
Subject: Re: [PATCH 1/5] x86, nmi:  Add new nmi type 'external'
Date: Fri, 9 May 2014 09:10:50 +0200	[thread overview]
Message-ID: <20140509071050.GA19751@gmail.com> (raw)
In-Reply-To: <20140508175247.GA39568@redhat.com>

* Don Zickus <dzickus@redhat.com> wrote:

> On Thu, May 08, 2014 at 07:35:01PM +0200, Ingo Molnar wrote:
> > 
> > * Don Zickus <dzickus@redhat.com> wrote:
> > 
> > > > > Again, I don't have a solution to juggle between PMI performance 
> > > > > and reliable delivery.  We could do away with the spinlocks and 
> > > > > go back to single cpu delivery (like it used to be).  Then 
> > > > > devise a mechanism to switch delivery to another cpu upon 
> > > > > hotplug.
> > > > > 
> > > > > Thoughts?
> > > > 
> > > > I'd say we should do a delayed timer that makes sure that all 
> > > > possible handlers are polled after an NMI is triggered, but never 
> > > > at a high rate.
> > > 
> > > Hmm, I was thinking about it and wanted to avoid a poll as I hear 
> > > complaints here and there about the nmi_watchdog constantly wasting 
> > > power cycles with its polling.
> > 
> > But the polling would only happen if there's NMI traffic, so that's 
> > fine. So as long as polling stops some time after the last PMI use, 
> > it's a good solution.
> 
> So you are thinking an NMI comes in, kicks off a delayed timer for 
> say 10ms.  The timer fires, rechecks the NMI for missed events and 
> then stops? If another NMI happens before the timer fires, just kick 
> the timer again?
> 
> Something like that?

Yeah, exactly, using delayed IRQ work for that or so.

This would allow us to 'optimistic' processing of NMI events: the 
first handler that manages to do any work causes a return. No need to 
make a per handler distinction, etc.

It would generally be pretty robust and would possibly be a natural 
workaround for 'stuck PMU' type of bugs as well.

[ As long as it does not result in spurious 'dazed and confused' 
  messages :-) ]

Thanks,

	Ingo