Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 26 Oct 2022 14:58:34 +0200
From:   "Jason A. Donenfeld" <Jason@zx2c4.com>
To:     Kees Cook <keescook@chromium.org>, cocci@inria.fr
Cc:     Linus Torvalds <torvalds@linux-foundation.org>,
        Alexey Dobriyan <adobriyan@gmail.com>,
        akpm@linux-foundation.org, linux-kernel@vger.kernel.org,
        mm-commits@vger.kernel.org, masahiroy@kernel.org,
        gregkh@linuxfoundation.org, andriy.shevchenko@linux.intel.com,
        Stephen Rothwell <sfr@canb.auug.org.au>
Subject: Re: [PATCH -mm] -funsigned-char, x86: make struct
 p4_event_bind::cntr signed array
Message-ID: <Y1ku+jfRAyezq6Nz@zx2c4.com>
References: <20221020000356.177CDC433C1@smtp.kernel.org>
 <Y1EZuQcO8UoN91cX@localhost.localdomain>
 <CAHmME9prEhkHqQmtDGCSFunNnxiKdE_8FHKiksyqebUN63U81Q@mail.gmail.com>
 <CAHk-=whFow9Wd6C8htoRUt5wXbwf1i_qbuArBbhXOPqYsTFvtw@mail.gmail.com>
 <CAHmME9qBZqTd0D_gr8nE+DUzCrC0fxZNZK=7u+21jbgtFgAJBg@mail.gmail.com>
 <CAHk-=wjZDC9o8iwF+bU91Hx40HjGOpMui+VoFCDJkaGCu=rG4A@mail.gmail.com>
 <202210201151.ECC19BC97A@keescook>
 <Y1iSYddi3BpP8gvf@zx2c4.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <Y1iSYddi3BpP8gvf@zx2c4.com>
Precedence: bulk

On Wed, Oct 26, 2022 at 03:50:25AM +0200, Jason A. Donenfeld wrote:
> The traditional objdump comparison does work, though. It produces a good

Another thing that appears to work well is just using Coccinelle
scripts. I've had some success just scrolling through the results of:

    @@
    char c;
    expression E;
    @@
    (
    * E > c
    |
    * E >= c
    |
    * E < c
    |
    * E <= c
    )

That also triggers on explicitly signed chars, and examining those
reveals that quite a bit of code in the tree already does do the right
thing, which is good.

From looking at this and objdump output, it looks like most naked-char
usage that isn't for strings is actually already assuming it's unsigned,
using it as a byte. I'll continue to churn, and I'm sure I'll miss a few
things here and there, but all and all, I don't think this is looking as
terrible as I initially feared.

I'm CC'ing the Coccinelle people to see if they have any nice ideas on
improvements. Specifically, the thing we're trying to identify is:

  - Usage of vanilla `char`, without a `signed` or `unsigned` qualifier,
    where:
  - It's not being used for characters; and
  - It's doing something that assumes it is signed, such as various
    types of comparisons or decrements.

LWN wrote a summary of the general problem, in case that helps describe
what would be useful: https://lwn.net/SubscriberLink/911914/f90c2ed1af23cbc4/

Any nice Cocci tricks for this?

Jason