linux-kernel - Re: [LKP] [fsnotify] 60f7ed8c7c: will-it-scale.per_thread

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Tue, 2 Oct 2018 17:49:27 +0300
From:   Amir Goldstein <amir73il@...il.com>
To:     Jan Kara <jack@...e.cz>
Cc:     rong.a.chen@...el.com, linux-kernel <linux-kernel@...r.kernel.org>,
        Stephen Rothwell <sfr@...b.auug.org.au>, LKP <lkp@...org>
Subject: Re: [LKP] [fsnotify] 60f7ed8c7c: will-it-scale.per_thread_ops -5.9% regression

On Mon, Oct 1, 2018 at 12:52 PM Amir Goldstein <amir73il@...il.com> wrote:
>
> On Mon, Oct 1, 2018 at 12:32 PM Jan Kara <jack@...e.cz> wrote:
> >
> > On Sun 30-09-18 12:00:46, Amir Goldstein wrote:
> [...]
> > > > commit:
> > > >   1e6cb72399 ("fsnotify: add super block object type")
> > > >   60f7ed8c7c ("fsnotify: send path type events to group with super block marks")
> > > >
> > >
> > > I have to admit this looks strange.
> > > All this commit does is dereference mnt->mnt.mnt_sb and then
> > > sb->s_fsnotify_mask/sb->s_fsnotify_marks to find that they are zero.
> > > AFAICT there should be no extra contention added by this commit and it's
> > > hard to believe that parallel unlink workload would suffer from this change.
> >
> > Well, it could be those additional fetches of
> > sb->s_fsnotify_mask/sb->s_fsnotify_marks if they happen to be cache cold.
> > Or it could be just code layout differences (i.e., compiler is not able to
> > optimize resulting code as good or the code layout just happens to align
> > with cache lines in a wrong way or something like that). Anyway, without
> > being able to reproduce this and do detailed comparison of perf profiles I
> > don't think we'll be able to tell.
> >
>

On my test machine, I couldn't measure a difference in results between the
two commits, but it may be because my test machine is not strong enough to
drive the test. It only has 4 cores and so I could only reach ~1/4 of
the workload
reported by 0day robot.

I tested both 16-thread and 4-thread jobs. In both cases the measured difference
between the two commits was less than 0.5% and within the stddev margins.

I will see if I can get my hands on a stronger test machine.

Thanks,
Amir.