Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20211005105905.1994700-1-elver@google.com> <20211005105905.1994700-17-elver@google.com>
 <YVw+4McyFdvU7ZED@hirez.programming.kicks-ass.net>
In-Reply-To: <YVw+4McyFdvU7ZED@hirez.programming.kicks-ass.net>
From:   Marco Elver <elver@google.com>
Date:   Tue, 5 Oct 2021 14:16:55 +0200
Message-ID: <CANpmjNO6H2imqsGaLYqimm0POvqA65Pd3OYji-QzONMn=Ht6Og@mail.gmail.com>
Subject: Re: [PATCH -rcu/kcsan 16/23] locking/atomics, kcsan: Add
 instrumentation for barriers
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     "Paul E . McKenney" <paulmck@kernel.org>,
        Alexander Potapenko <glider@google.com>,
        Boqun Feng <boqun.feng@gmail.com>,
        Borislav Petkov <bp@alien8.de>,
        Dmitry Vyukov <dvyukov@google.com>,
        Ingo Molnar <mingo@kernel.org>,
        Josh Poimboeuf <jpoimboe@redhat.com>,
        Mark Rutland <mark.rutland@arm.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Waiman Long <longman@redhat.com>,
        Will Deacon <will@kernel.org>, kasan-dev@googlegroups.com,
        linux-arch@vger.kernel.org, linux-doc@vger.kernel.org,
        linux-kbuild@vger.kernel.org, linux-kernel@vger.kernel.org,
        linux-mm@kvack.org, x86@kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, 5 Oct 2021 at 14:03, Peter Zijlstra <peterz@infradead.org> wrote:
>
> On Tue, Oct 05, 2021 at 12:58:58PM +0200, Marco Elver wrote:
> > @@ -59,6 +60,7 @@ atomic_add(int i, atomic_t *v)
> >  static __always_inline int
> >  atomic_add_return(int i, atomic_t *v)
> >  {
> > +     kcsan_mb();
> >       instrument_atomic_read_write(v, sizeof(*v));
> >       return arch_atomic_add_return(i, v);
> >  }
>
> This and others,.. is this actually correct? Should that not be
> something like:
>
>         kscan_mb();
>         instrument_atomic_read_write(...);
>         ret = arch_atomic_add_return(i, v);
>         kcsan_mb();
>         return ret;
>
> ?

In theory, yes, but right now it's redundant.

Because right now KCSAN only models "buffering", and no "prefetching".
So there's no way that a later instruction would be reordered before
this point. And atomic accesses are never considered for reordering,
so it's also impossible that it would  be reordered later.

Each kcsan_mb() is a call, so right now it makes sense to just have 1
call to be a bit more efficient.