From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S262584AbULPCP4 (ORCPT ); Wed, 15 Dec 2004 21:15:56 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S262566AbULPCMh (ORCPT ); Wed, 15 Dec 2004 21:12:37 -0500 Received: from fsmlabs.com ([168.103.115.128]:18345 "EHLO fsmlabs.com") by vger.kernel.org with ESMTP id S262582AbULPCLD (ORCPT ); Wed, 15 Dec 2004 21:11:03 -0500 Date: Wed, 15 Dec 2004 19:10:17 -0700 (MST) From: Zwane Mwaikambo To: Alan Cox cc: Linus Torvalds , Ingo Molnar , Lee Revell , Andrea Arcangeli , Manfred Spraul , George Anzinger , dipankar@in.ibm.com, ganzinger@mvista.com, lkml , Andrew Morton , Andi Kleen Subject: Re: [patch, 2.6.10-rc3] safe_hlt() & NMIs In-Reply-To: <1103157476.3585.33.camel@localhost.localdomain> Message-ID: References: <41BB2108.70606@colorfullife.com> <41BB25B2.90303@mvista.com> <41BC0854.4010503@colorfullife.com> <20041212093714.GL16322@dualathlon.random> <41BC1BF9.70701@colorfullife.com> <20041212121546.GM16322@dualathlon.random> <1103060437.14699.27.camel@krustophenia.net> <20041214222307.GB22043@elte.hu> <20041214224706.GA26853@elte.hu> <1103157476.3585.33.camel@localhost.localdomain> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Thu, 16 Dec 2004, Alan Cox wrote: > On Maw, 2004-12-14 at 23:09, Linus Torvalds wrote: > > On Tue, 14 Dec 2004, Ingo Molnar wrote: > > Now that you mention it, I have this dim memory of the one-instruction > > "sti-shadow" actually disabling NMI's (and debug traps) too. The CPU > > literally doesn't test for async events following "sti". > > > > Or maybe that was "mov->ss". That one also has that strange "black hole" > > for one instruction. > > The mov to ss one is a bit more magic than that however. If you write > 3Gb of mov->ss into memory (ie about 64 pages to thrash the cache and > slow it plus mmap repeatedly) and run it you don't get a vastly long irq > delay at least on intel, not tried the others. Might this be because you can't rely on interrupt suppression for back to back suppressing instructions?