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;
MIME-Version: 1.0
References: <20230517181353.381073-1-kursad.oney@broadcom.com>
In-Reply-To: <20230517181353.381073-1-kursad.oney@broadcom.com>
From:   Ard Biesheuvel <ardb@kernel.org>
Date:   Thu, 3 Aug 2023 15:59:50 +0200
Message-ID: <CAMj1kXH0HJg2g_6TspGhMVkWLo4wGCmXXnno2ty-sbDwu9ft_A@mail.gmail.com>
Subject: Re: [PATCH] ARM: memset: cast the constant byte to unsigned char
To:     Kursad Oney <kursad.oney@broadcom.com>,
        Linus Walleij <linus.walleij@linaro.org>
Cc:     linux-arm-kernel@lists.infradead.org,
        BCM Kernel Feedback <bcm-kernel-feedback-list@broadcom.com>,
        Russell King <linux@armlinux.org.uk>,
        linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Wed, 17 May 2023 at 20:14, Kursad Oney <kursad.oney@broadcom.com> wrote:
>
> memset() description in ISO/IEC 9899:1999 (and elsewhere) says:
>
>         The memset function copies the value of c (converted to an
>         unsigned char) into each of the first n characters of the
>         object pointed to by s.
>
> The kernel's arm32 memset does not cast c to unsigned char. This results
> in the following code to produce erroneous output:
>
>         char a[128];
>         memset(a, -128, sizeof(a));
>
> This is because gcc will generally emit the following code before
> it calls memset() :
>
>         mov   r0, r7
>         mvn   r1, #127        ; 0x7f
>         bl    00000000 <memset>
>
> r1 ends up with 0xffffff80 before being used by memset() and the
> 'a' array will have -128 once in every four bytes while the other
> bytes will be set incorrectly to -1 like this (printing the first
> 8 bytes) :
>
>         test_module: -128 -1 -1 -1
>         test_module: -1 -1 -1 -128
>
> The change here is to 'and' r1 with 255 before it is used.
>
> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> Signed-off-by: Kursad Oney <kursad.oney@broadcom.com>
>
> ---
>
>  arch/arm/lib/memset.S | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/arch/arm/lib/memset.S b/arch/arm/lib/memset.S
> index d71ab61430b2..de75ae4d5ab4 100644
> --- a/arch/arm/lib/memset.S
> +++ b/arch/arm/lib/memset.S
> @@ -17,6 +17,7 @@ ENTRY(__memset)
>  ENTRY(mmioset)
>  WEAK(memset)
>  UNWIND( .fnstart         )
> +       and     r1, r1, #255            @ cast to unsigned char
>         ands    r3, r0, #3              @ 1 unaligned?
>         mov     ip, r0                  @ preserve r0 as return value
>         bne     6f                      @ 1

Yes, this is clearly a bug. The value in R1 is expanded to 32 bits like this

1:      orr     r1, r1, r1, lsl #8
        orr     r1, r1, r1, lsl #16

which assumes that the upper bytes are 0x0, which they are not in this case.


Reviewed-by: Ard Biesheuvel <ardb@kernel.org>