linux-kernel - Re: [PATCH v2] scripts: coccinelle: check for !(un)?likely usage

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: <a327be9e-cd62-6c22-a9b0-ba0f9b295737@rasmusvillemoes.dk>
Date:   Fri, 30 Aug 2019 10:06:39 +0200
From:   Rasmus Villemoes <linux@...musvillemoes.dk>
To:     Denis Efremov <efremov@...ux.com>,
        Julia Lawall <julia.lawall@...6.fr>
Cc:     linux-kernel@...r.kernel.org, cocci@...teme.lip6.fr,
        Gilles Muller <Gilles.Muller@...6.fr>,
        Nicolas Palix <nicolas.palix@...g.fr>,
        Michal Marek <michal.lkml@...kovi.net>,
        Markus Elfring <Markus.Elfring@....de>,
        Joe Perches <joe@...ches.com>
Subject: Re: [PATCH v2] scripts: coccinelle: check for !(un)?likely usage

On 30/08/2019 08.56, Denis Efremov wrote:
> 
> 
> On 30.08.2019 03:42, Julia Lawall wrote:
>>
>>
>> On Thu, 29 Aug 2019, Denis Efremov wrote:
>>
>>> On 8/29/19 8:10 PM, Denis Efremov wrote:
>>>> This patch adds coccinelle script for detecting !likely and
>>>> !unlikely usage. These notations are confusing. It's better
>>>> to replace !likely(x) with unlikely(!x) and !unlikely(x) with
>>>> likely(!x) for readability.
>>>
>>> I'm not sure that this rule deserves the acceptance.
>>> Just to want to be sure that "!unlikely(x)" and "!likely(x)"
>>> are hard-readable is not only my perception and that they
>>> become more clear in form "likely(!x)" and "unlikely(!x)" too.
>>
>> Is likely/unlikely even useful for anything once it is a subexpression?
>>> julia
>>
> 
> Well, as far as I understand it,

Yes, and it could in fact make sense in cases like

  if (likely(foo->bar) && unlikely(foo->bar->baz)) {
     do_stuff_with(foo->bar->baz);
     do_more_stuff();
  }

which the compiler could then compile as (of course actual code
generation is always much more complicated due to things in the
surrounding code)

load foo->bar;
test bar;
if 0 goto skip;
load bar->baz;
test baz;
if !0 goto far_away;
skip:
  ....

so in the normal flow, neither branch is taken. If instead one wrote
unlikely(foo->bar && foo->bar->baz), gcc doesn't really know why we
expect the whole conjuntion to turn out false, so it could compile this
as a jump when foo->bar turns out non-zero - i.e., in the normal case,
we'd end up jumping.

But as far as !(un)likely(), I agree that it's easier to read as a human
if the ! operator is moved inside (and the "un" prefix stripped/added).
Whether it deserves a cocci script I don't know.

Rasmus