linux-kernel - Re: [PATCH] schedule: use unlikely()

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

Message-ID: <alpine.LRH.2.02.1711300200470.25502@file01.intranet.prod.int.rdu2.redhat.com>
Date:   Thu, 30 Nov 2017 02:04:01 -0500 (EST)
From:   Mikulas Patocka <mpatocka@...hat.com>
To:     Greg KH <gregkh@...uxfoundation.org>
cc:     Ingo Molnar <mingo@...hat.com>,
        Peter Zijlstra <peterz@...radead.org>,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH] schedule: use unlikely()



On Tue, 28 Nov 2017, Greg KH wrote:

> On Mon, Nov 27, 2017 at 07:05:22PM -0500, Mikulas Patocka wrote:
> > 
> > 
> > On Sat, 25 Nov 2017, Greg KH wrote:
> > 
> > > On Mon, Nov 13, 2017 at 02:00:45PM -0500, Mikulas Patocka wrote:
> > > > A small patch for schedule(), so that the code goes straght in the common
> > > > case.
> > > > 
> > > > Signed-off-by: Mikulas Patocka <mpatocka@...hat.com>
> > > 
> > > Was this a measurable difference?  If so, great, please provide the
> > > numbers and how you tested in the changelog.  If it can't be measured,
> > > then it is not worth it to add these markings
> > 
> > It is much easier to make microoptimizations (such as using likely() and 
> > unlikely()) than to measure their effect.
> > 
> > If a programmer were required to measure performance every time he uses 
> > likely() or unlikely() in his code, he wouldn't use them at all.
> 
> If you can not measure it, you should not use it.  You are forgetting
> about the testing that was done a few years ago that found that some
> huge percentage (80? 75? 90?) of all of these markings were wrong and
> harmful or did absolutely nothing.

The whole kernel has 19878 likely/unlikely tags.

Do you have benchmark proving efficiency for each of them? :-)

Mikulas