linux-kernel - Re: [PATCH, RESEND 1/2] dvb-frontends: fix i2c access helpers for KASAN

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: <CAK8P3a0dp0S8h0pV+1mexD=-LdRRW8D55tLmx5x9usCQkzNTqw@mail.gmail.com>
Date:   Thu, 30 Nov 2017 15:06:15 +0100
From:   Arnd Bergmann <arnd@...db.de>
To:     Mauro Carvalho Chehab <mchehab@...nel.org>
Cc:     "# 3.4.x" <stable@...r.kernel.org>, Sergey Kozlov <serjk@...up.ru>,
        Abylay Ospan <aospan@...up.ru>,
        Daniel Scheller <d.scheller@....net>,
        Alexey Dobriyan <adobriyan@...il.com>,
        Masanari Iida <standby24x7@...il.com>,
        Jiri Kosina <jkosina@...e.cz>,
        Randy Dunlap <rdunlap@...radead.org>,
        Sakari Ailus <sakari.ailus@...ux.intel.com>,
        Linux Media Mailing List <linux-media@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH, RESEND 1/2] dvb-frontends: fix i2c access helpers for KASAN

On Thu, Nov 30, 2017 at 1:49 PM, Mauro Carvalho Chehab
<mchehab@...nel.org> wrote:
>> Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81715
>> Signed-off-by: Arnd Bergmann <arnd@...db.de>
>> ---
>> I'm undecided here whether there should be a comment pointing
>> to PR81715 for each file that the bogus local variable workaround
>> to prevent it from being cleaned up again. It's probably not
>> necessary since anything that causes actual problems would also
>> trigger a build warning.

>
> This kind of sucks, and it is completely unexpected... why val is
> so special that it would require this kind of hack?

It's explained in the gcc bug report: basically gcc always skipped
one optimization on inline function arguments that it does on
normal variables. Without KASAN and asan-stack, we didn't
notice because the impact was fairly small, but I ended up finally
getting to the bottom of it in September, and it finally got fixed.

I had an older version of the patch that was much more invasive
before we understood what exactly is happening, see
https://lkml.org/lkml/2017/3/2/484

> Also, there's always a risk of someone see it and decide to
> simplify the code, returning it to the previous state.
>
> So, if we're willing to do something like that, IMHO, we should have
> some macro that would document it, and fall back to the direct
> code if the compiler is not gcc 5, 6 or 7.

Older compilers are also affected and will produce better code
with my change, the difference is just smaller without asan-stack
(added ion gcc-5) is disabled, since that increases the stack
space used by each variable to (IIRC) 32 bytes.

The fixed gcc-8 produces identical code with and without my
change.

I don't think that a macro would help here at all, but if you
prefer, I could add a link to that gcc bug in each function that
has the problem.

         Arnd