From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-ee0-f50.google.com (mail-ee0-f50.google.com [74.125.83.50]) by kanga.kvack.org (Postfix) with ESMTP id B02946B0036 for ; Thu, 15 May 2014 08:23:46 -0400 (EDT) Received: by mail-ee0-f50.google.com with SMTP id e51so598814eek.37 for ; Thu, 15 May 2014 05:23:46 -0700 (PDT) Received: from mx1.redhat.com (mx1.redhat.com. [209.132.183.28]) by mx.google.com with ESMTP id h6si4058749eew.101.2014.05.15.05.23.44 for ; Thu, 15 May 2014 05:23:45 -0700 (PDT) From: Naoya Horiguchi Subject: Re: [PATCH] mm/memory-failure.c: fix memory leak by race between poison and unpoison Date: Thu, 15 May 2014 08:23:10 -0400 Message-Id: <5374b1d1.86300f0a.4a16.65ffSMTPIN_ADDED_BROKEN@mx.google.com> In-Reply-To: <1400124866.26173.19.camel@cyc> References: <1400080891-5145-1-git-send-email-n-horiguchi@ah.jp.nec.com> <1400124866.26173.19.camel@cyc> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: quoted-printable Content-Disposition: inline Sender: owner-linux-mm@kvack.org List-ID: To: soldier.cyc81@gmail.com Cc: Andrew Morton , Andi Kleen , Wu Fengguang , linux-kernel@vger.kernel.org, linux-mm@kvack.org On Thu, May 15, 2014 at 11:34:26AM +0800, cyc wrote: > =E5=9C=A8 2014-05-14=E4=B8=89=E7=9A=84 11:21 -0400=EF=BC=8CNaoya Horigu= chi=E5=86=99=E9=81=93=EF=BC=9A > > When a memory error happens on an in-use page or (free and in-use) hu= gepage, > > the victim page is isolated with its refcount set to one. When you tr= y to > > unpoison it later, unpoison_memory() calls put_page() for it twice in= order to > > bring the page back to free page pool (buddy or free hugepage list.) > > However, if another memory error occurs on the page which we are unpo= isoning, > > memory_failure() returns without releasing the refcount which was inc= remented > > in the same call at first, which results in memory leak and unconsist= ent > > num_poisoned_pages statistics. This patch fixes it. > = > We assume that a new memory error occurs on the hugepage which we are > unpoisoning. = > = > A unpoisoned B poisoned C = > hugepage: |---------------+++++++++++++++++| > = > There are two cases, so shown. > 1. the victim page belongs to A-B, the memory_failure will be blocked= > by lock_page() until unlock_page() invoked by unpoison_memory(). No. memory_failure() set PageHWPoison at first before taking page lock. This is a design choice based on the idea that we need detect errors ASAP= . What happens in this race is like below: CPU 0 (poison) CPU 1 (unpoison) lock_page TestSetPageHWPoison TestClearPageHWPoison lock_page (wait) unlock_page check PageHWPoison printk("just unpoisoned") > 2. the victim page belongs to B-C, the memory_failure() will return > very soon at the beginning of this function. Right. Thanks, Naoya Horiguchi -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: email@kvack.org