From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753612AbaBPTnd (ORCPT ); Sun, 16 Feb 2014 14:43:33 -0500 Received: from one.firstfloor.org ([193.170.194.197]:48003 "EHLO one.firstfloor.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753588AbaBPTnc (ORCPT ); Sun, 16 Feb 2014 14:43:32 -0500 Date: Sun, 16 Feb 2014 20:43:30 +0100 From: Andi Kleen To: Peter Zijlstra Cc: Andi Kleen , mingo@kernel.org, eranian@google.com, linux-kernel@vger.kernel.org, Markus Metzger , Andi Kleen Subject: Re: [PATCH] perf, nmi: fix unknown NMI warning Message-ID: <20140216194330.GE32005@two.firstfloor.org> References: <1392425048-5309-1-git-send-email-andi@firstfloor.org> <20140215095843.GJ14089@laptop.programming.kicks-ass.net> <20140216183850.GD32005@two.firstfloor.org> <20140216192312.GM14089@laptop.programming.kicks-ass.net> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20140216192312.GM14089@laptop.programming.kicks-ass.net> User-Agent: Mutt/1.5.20 (2009-06-14) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org The best APIC documentation are the old data sheets for the external APIC chips. I don't know if they cover things in such detail. > In this case the latter NMI will actually have an overflow state to > process so it's not a spurious NMI. But we cannot distinguish it right? The spurious detector would trigger in any case. > > > And if we're in a state that PMIs get re-raised quickly, we should either > > regulate the period down or start throttling. > > It could be a different counter; where both run at 'normal' periods but > just near miss each other by accident. That's true. It would be only a problem if they somehow become synchronized that this happens very commonly. The usual defense against things like that is to add a little randomization (I remember Stephane had a patch for that some time ago). Also I believe it helps to have the periods be prime numbers. But right now don't have any evidence it's a real problem. I presume there's enough noise on a typical setup that any such states disappear again quickly enough. -Andi