Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH] lib/strscpy: remove word-at-a-time optimization.
To:     Andrey Ryabinin <aryabinin@virtuozzo.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Linus Torvalds <torvalds@linux-foundation.org>
CC:     <linux-kernel@vger.kernel.org>, Kees Cook <keescook@chromium.org>,
        Eryu Guan <eguan@redhat.com>,
        Alexander Potapenko <glider@google.com>,
        Chris Metcalf <metcalf@alum.mit.edu>,
        David Laight <David.Laight@ACULAB.COM>,
        Dmitry Vyukov <dvyukov@google.com>, <stable@vger.kernel.org>
References: <20180109163745.3692-1-aryabinin@virtuozzo.com>
From:   Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Message-ID: <317d8c15-b4af-c0d6-31c9-30e311327b3f@prevas.dk>
Date:   Wed, 24 Jan 2018 09:47:41 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
 Thunderbird/52.5.0
MIME-Version: 1.0
In-Reply-To: <20180109163745.3692-1-aryabinin@virtuozzo.com>
Content-Type: text/plain; charset="utf-8"
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 2018-01-09 17:37, Andrey Ryabinin wrote:
> strscpy() performs the word-at-a-time optimistic reads. So it may
> may access the memory past the end of the object, which is perfectly fine
> since strscpy() doesn't use that (past-the-end) data and makes sure the
> optimistic read won't cross a page boundary.
> 
> But KASAN doesn't know anything about that so it will complain.
> There are several possible ways to address this issue, but none
> are perfect. See https://lkml.kernel.org/r/9f0a9cf6-51f7-cd1f-5dc6-6d510a7b8ec4@virtuozzo.com
> 
> It seems the best solution is to simply disable word-at-a-time
> optimization. My trivial testing shows that byte-at-a-time
> could be up to x4.3 times slower than word-at-a-time.
> It may seems like a lot, but it's actually ~1.2e-10 sec per symbol vs
> ~4.8e-10 sec per symbol on modern hardware. And we don't use strscpy()
> in a performance critical paths to copy large amounts of data,
> so it shouldn't matter anyway.
> 
> Fixes: 30035e45753b7 ("string: provide strscpy()")
> Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
> Cc: <stable@vger.kernel.org>
>

Acked-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>

Your microbenchmark even favours word-at-a-time slightly, since in
practice I think at least one of src or dst will be unaligned a lot of
the time, and while x86 may HAVE_EFFICIENT_UNALIGNED_ACCESS, it's still
a little more expensive than doing aligned access. And since strscpy is
not called that often, I expect some of the ~300 bytes of instruction
cache it occupies can be put to better use elsewhere.

Rasmus