From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-qt1-f195.google.com ([209.85.160.195]:43733 "EHLO mail-qt1-f195.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726137AbfAIVNQ (ORCPT ); Wed, 9 Jan 2019 16:13:16 -0500 Received: by mail-qt1-f195.google.com with SMTP id i7so9968277qtj.10 for ; Wed, 09 Jan 2019 13:13:15 -0800 (PST) Message-ID: <1547068393.6911.3.camel@lca.pw> Subject: Re: [PATCH] xfs: silence lockdep false positives when freezing From: Qian Cai Date: Wed, 09 Jan 2019 16:13:13 -0500 In-Reply-To: <20190109210111.GZ4205@dastard> References: <20190106225639.GU4205@dastard> <20190109205329.2486-1-cai@lca.pw> <20190109210111.GZ4205@dastard> Content-Type: text/plain; charset="UTF-8" Mime-Version: 1.0 Content-Transfer-Encoding: 8bit Sender: linux-xfs-owner@vger.kernel.org List-ID: List-Id: xfs To: Dave Chinner Cc: darrick.wong@oracle.com, dchinner@redhat.com, peterz@infradead.org, bfoster@redhat.com, hch@lst.de, linux-xfs@vger.kernel.org, linux-kernel@vger.kernel.org On Thu, 2019-01-10 at 08:01 +1100, Dave Chinner wrote: > On Wed, Jan 09, 2019 at 03:53:29PM -0500, Qian Cai wrote: > > Easy to reproduce: > > > > 1. run LTP oom02 workload to let kswapd acquire this locking order: > >    fs_reclaim -> sb_internal. > > > >  # grep -i fs_reclaim -C 3 /proc/lockdep_chains | grep -C 5 sb_internal > > [00000000826b9172] &type->s_umount_key#27 > > [000000005fa8b2ac] sb_pagefaults > > [0000000033f1247e] sb_internal > > [000000009e9a9664] fs_reclaim > > > > 2. freeze XFS. > >   # fsfreeze -f /home > > > > Dave mentioned that this is due to a lockdep limitation - "IOWs, this is > > a false positive, caused by the fact that xfs_trans_alloc() is called > > from both above and below memory reclaim as well as within /every level/ > > of freeze processing. Lockdep is unable to describe the staged flush > > logic in the freeze process that prevents deadlocks from occurring, and > > hence we will pretty much always see false positives in the freeze > > path....". Hence, just temporarily disable lockdep in that path. > > NACK. Turning off lockdep is not a solution, it just prevents > lockdep from finding and reporting real issues. > Well, it is a trade-off. It is turned on right after that path. All those false positives leave unfixed are also going to render lockdep less useful.