linux-kernel - Re: RCU vs data

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: <YMyC0iux0wKzc1JG@hirez.programming.kicks-ass.net>
Date:   Fri, 18 Jun 2021 13:26:10 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     Marco Elver <elver@...gle.com>
Cc:     Paul McKenney <paulmck@...nel.org>, linux-kernel@...r.kernel.org
Subject: Re: RCU vs data_race()

On Fri, Jun 18, 2021 at 10:59:26AM +0200, Marco Elver wrote:
> On Fri, Jun 18, 2021 at 10:24AM +0200, Peter Zijlstra wrote:
> > Hi Paul,
> > 
> > Due to a merge conflict I had to look at some recent RCU code, and I saw
> > you went a little overboard with data_race(). How's something like the
> > below look to you?
> 
> I commented below. The main thing is just using the __no_kcsan function
> attribute if it's only about accesses within the function (and not
> also about called functions elsewhere).
> 
> Using the attribute also improves performance slightly (not that it
> matters much in a KCSAN-enabled kernel) due to no instrumentation.

Aah yes ofcourse! Much better still.

> > The idea being that we fundamentally don't care about data races for
> > debug/error condition prints, so marking every single variable access is
> > just clutter.
> 
> Having data_race() around the pr_* helpers seems reasonable, if you
> worry about future unnecessary markings that might pop up due to them.

Right, so I did them on WARN and higher, figuring that those really only
happen when there's smoething wrong and we're way past caring about
data races. Paul has a few pr_info() users that are heavy on
data_race(), but your __no_kcsan attribute suggestion helps with that.