Content-Type: multipart/mixed; boundary="===============5374090044488641783==" MIME-Version: 1.0 From: Al Viro To: lkp@lists.01.org Subject: Re: [rcuwalk] e2a728c2785: -29.4% aim7.2000.jobs-per-min, +71% turbostat.Pkg_W Date: Sat, 22 Mar 2014 13:54:12 +0000 Message-ID: <20140322135412.GF18016@ZenIV.linux.org.uk> In-Reply-To: <20140322115636.GA14400@localhost> List-Id: --===============5374090044488641783== Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable On Sat, Mar 22, 2014 at 07:56:36PM +0800, Fengguang Wu wrote: > Hi Al, > = > We noticed the below changes on > = > git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git untested-mnt_h= ash > commit e2a728c2785d6aab45d1a6d2c61e986d385c30bf ("rcuwalk: recheck mount_= lock after mountpoint crossing attempts") *doh* Sorry about that - what happens is that __follow_mount_rcu() does that check every time it's called, not on every negative lookup from __lookup_mnt(). IOW, it's a massive overkill - most of the time we don't call __lookup_mnt() at all, due to d_mountpoint() being false, or get a positive from it. Fixed and force-pushed; should get much better now... --===============5374090044488641783==--