From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932161AbcGAXEL (ORCPT ); Fri, 1 Jul 2016 19:04:11 -0400 Received: from gate.crashing.org ([63.228.1.57]:34092 "EHLO gate.crashing.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932068AbcGAXEJ (ORCPT ); Fri, 1 Jul 2016 19:04:09 -0400 Message-ID: <1467412092.7422.56.camel@kernel.crashing.org> Subject: Re: [PATCH 0/4] [RFC][v4] Workaround for Xeon Phi PTE A/D bits erratum From: Benjamin Herrenschmidt To: Dave Hansen , linux-kernel@vger.kernel.org Cc: x86@kernel.org, linux-mm@kvack.org, torvalds@linux-foundation.org, akpm@linux-foundation.org, bp@alien8.de, ak@linux.intel.com, mhocko@suse.com Date: Sat, 02 Jul 2016 08:28:12 +1000 In-Reply-To: <20160701174658.6ED27E64@viggo.jf.intel.com> References: <20160701174658.6ED27E64@viggo.jf.intel.com> Content-Type: text/plain; charset="UTF-8" X-Mailer: Evolution 3.20.3 (3.20.3-1.fc24.1) Mime-Version: 1.0 Content-Transfer-Encoding: 8bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Fri, 2016-07-01 at 10:46 -0700, Dave Hansen wrote: > The Intel(R) Xeon Phi(TM) Processor x200 Family (codename: Knights > Landing) has an erratum where a processor thread setting the Accessed > or Dirty bits may not do so atomically against its checks for the > Present bit.  This may cause a thread (which is about to page fault) > to set A and/or D, even though the Present bit had already been > atomically cleared. Interesting.... I always wondered where in the Intel docs did it specify that present was tested atomically with setting of A and D ... I couldn't find it. Isn't there a more fundamental issue however that you may actually lose those bits ? For example if we do an munmap, in zap_pte_range() We first exchange all the PTEs with 0 with ptep_get_and_clear_full() and we then transfer D that we just read into the struct page. We rely on the fact that D will never be set again, what we go it a "final" D bit. IE. We rely on the fact that a processor either: - Has a cached PTE in its TLB with D set, in which case it can still write to the page until we flush the TLB or - Doesn't have a cached PTE in its TLB with D set and so will fail to do so due to the atomic P check, thus never writing. With the errata, don't you have a situation where a processor in the second category will write and set D despite P having been cleared (due to the race) and thus causing us to miss the transfer of that D to the struct page and essentially completely miss that the physical page is dirty ? (Leading to memory corruption). > If the PTE is used for storing a swap index or a NUMA migration index, > the A bit could be misinterpreted as part of the swap type.  The stray > bits being set cause a software-cleared PTE to be interpreted as a > swap entry.  In some cases (like when the swap index ends up being > for a non-existent swapfile), the kernel detects the stray value > and WARN()s about it, but there is no guarantee that the kernel can > always detect it. > > This patch changes the kernel to attempt to ignore those stray bits > when they get set.  We do this by making our swap PTE format > completely ignore the A/D bits, and also by ignoring them in our > pte_none() checks. > > Andi Kleen wrote the original version of this patch.  Dave Hansen > wrote the later ones.