Received-SPF: pass (google.com: domain of linux-crypto-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Florian Weimer <fweimer@redhat.com>
To:     "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc:     linux-kernel@vger.kernel.org, patches@lists.linux.dev,
        tglx@linutronix.de, linux-crypto@vger.kernel.org,
        linux-api@vger.kernel.org, x86@kernel.org,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>,
        Carlos O'Donell <carlos@redhat.com>,
        Arnd Bergmann <arnd@arndb.de>,
        Christian Brauner <brauner@kernel.org>
Subject: Re: [PATCH v11 3/4] random: introduce generic vDSO getrandom()
 implementation
References: <20221205020046.1876356-1-Jason@zx2c4.com>
        <20221205020046.1876356-4-Jason@zx2c4.com>
Date:   Mon, 05 Dec 2022 19:56:47 +0100
In-Reply-To: <20221205020046.1876356-4-Jason@zx2c4.com> (Jason A. Donenfeld's
        message of "Mon, 5 Dec 2022 03:00:45 +0100")
Message-ID: <878rjlr85s.fsf@oldenburg.str.redhat.com>
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/27.2 (gnu/linux)
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

* Jason A. Donenfeld:

> +retry_generation:
> +	/*
> +	 * @rng_info->generation must always be read here, as it serializes @st=
ate->key with the
> +	 * kernel's RNG reseeding schedule.
> +	 */
> +	current_generation =3D READ_ONCE(rng_info->generation);

> +		if (unlikely(READ_ONCE(state->generation) !=3D READ_ONCE(rng_info-

I'm pretty sure you need some sort of barrier here.  We have a similar
TM-lite construct in glibc ld.so for locating link maps by address, and
there the compiler performed reordering.

  _dl_find_object miscompilation on powerpc64le
  <https://sourceware.org/bugzilla/show_bug.cgi?id=3D28745>

I'm not familiar with READ_ONCE, but Documentation/atomic_t.txt suggests
it's a =E2=80=9Cregular LOAD=E2=80=9D, and include/asm-generic/rwonce.h con=
curs.
Likewise, the signal safety mechanism needs compiler barriers (signal
fences).

I'm also not sure how READ_ONCE realizes atomic 64-bit reads on 32-bit
platforms.  i386 can do them in user space via the FPU worst-case (if
the control word hasn't been corrupted).  CMPXCHG8B is not applicable
here because it's a read-only mapping.  Maybe add a comment at least
about that =E2=80=9Cstrong prevailing wind=E2=80=9D?

Thanks,
Florian