From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from mail-pf0-f196.google.com ([209.85.192.196]:36251 "EHLO mail-pf0-f196.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753404AbdEWCyl (ORCPT ); Mon, 22 May 2017 22:54:41 -0400 Received: by mail-pf0-f196.google.com with SMTP id n23so24085306pfb.3 for ; Mon, 22 May 2017 19:54:41 -0700 (PDT) Date: Mon, 22 May 2017 19:54:37 -0700 From: Eric Biggers Subject: Re: [PATCH] kvm-xfstests, gce-xfstests: improve recommended kernel config Message-ID: <20170523025437.GA606@zzz> References: <20170522172420.136541-1-ebiggers3@gmail.com> <20170523021938.biqavccr6bdyy5sc@thunk.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20170523021938.biqavccr6bdyy5sc@thunk.org> Sender: fstests-owner@vger.kernel.org To: Theodore Ts'o Cc: fstests@vger.kernel.org, Amir Goldstein , Eric Biggers List-ID: Hi Ted, On Mon, May 22, 2017 at 10:19:38PM -0400, Theodore Ts'o wrote: > On Mon, May 22, 2017 at 10:24:20AM -0700, Eric Biggers wrote: > > -CONFIG_LOCK_STAT=y > > -CONFIG_DEBUG_LOCKDEP=y > > Lockdep has *definitely* found bugs for us, and so I don't think we > want to turn it off. lockstat is a interesting case. It definitely > can cause performance degredations, but it also can very find > performance problems. > > I can definitely see us potentially looking for locking hotspots by > using a DAX configuration (or a /dev/ram configuration) using a > n1-highcpu-64 GCE VM. (Indeed, since Eric Whitney retired from HP, > we've lost our only "big iron" system for doing scalability testing.) > > The test appliance does currently export /proc/lock_stat via the www > interface, although we're currently not saving it in the results > tar.gz file. (That's an oversight.) > > So I have a slight preferene for leaving LOCK_STAT enabled, even if > we're not using most of the time. We're certainly not doing > performance benchmarks using this config, at any rate! > DEBUG_LOCKDEP is for debugging lockdep itself, not for turning on lockdep (that's CONFIG_PROVE_LOCKING). So I don't think it's needed. I'm fine with leaving on LOCK_STAT; I didn't know there was interest in it. Eric