lkml.org 
[lkml]   [2020]   [Dec]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] MIPS: boot: Fix unaligned access with CONFIG_MIPS_RAW_APPENDED_DTB
Hi Nick,

Le mer. 16 déc. 2020 à 18:08, Nick Desaulniers
<ndesaulniers@google.com> a écrit :
> On Wed, Dec 16, 2020 at 3:40 PM Paul Cercueil <paul@crapouillou.net>
> wrote:
>>
>> The compressed payload is not necesarily 4-byte aligned, at least
>> when
>> compiling with Clang. In that case, the 4-byte value appended to the
>> compressed payload that corresponds to the uncompressed kernel image
>> size must be read using get_unaligned_le().
>
> Should it be get_unaligned_le32()?

Indeed.

>>
>> This fixes Clang-built kernels not booting on MIPS (tested on a
>> Ingenic
>> JZ4770 board).
>>
>> Fixes: b8f54f2cde78 ("MIPS: ZBOOT: copy appended dtb to the end of
>> the kernel")
>> Cc: <stable@vger.kernel.org> # v4.7
>> Signed-off-by: Paul Cercueil <paul@crapouillou.net>
>
> Hi Paul, thanks for the patch (and for testing with Clang)!
> Alternatively, we could re-align __image_end to the next 4B multiple
> via:
>
> diff --git a/arch/mips/boot/compressed/ld.script
> b/arch/mips/boot/compressed/ld.script
> index 0ebb667274d6..349919eff5fb 100644
> --- a/arch/mips/boot/compressed/ld.script
> +++ b/arch/mips/boot/compressed/ld.script
> @@ -27,6 +27,7 @@ SECTIONS
> /* Put the compressed image here */
> __image_begin = .;
> *(.image)
> + . = ALIGN(4);
> __image_end = .;
> CONSTRUCTORS
> . = ALIGN(16);

Actually that would not work (I did try that), since the 4-byte size
appended to the compressed payload is inside the *(.image) section. The
code that appends it (in scripts/Makefile.lib, I think) doesn't seem to
take care about aligning it to a 4-byte offset. I have no idea why it
does with GCC and doesn't with Clang, and I have no idea why the
compressed payload's size isn't aligned either.

> The tradeoff being up to 3 wasted bytes of padding in the compressed
> image, vs fetching one value slower (assuming unaligned loads are
> slower than aligned loads MIPS, IDK). I doubt decompress_kernel is
> called repeatedly, so let's take the byte saving approach of yours by
> using unaligned loads!
>
> Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

Thanks.

Cheers,
-Paul

>> ---
>> arch/mips/boot/compressed/decompress.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/arch/mips/boot/compressed/decompress.c
>> b/arch/mips/boot/compressed/decompress.c
>> index c61c641674e6..47c07990432b 100644
>> --- a/arch/mips/boot/compressed/decompress.c
>> +++ b/arch/mips/boot/compressed/decompress.c
>> @@ -117,7 +117,7 @@ void decompress_kernel(unsigned long
>> boot_heap_start)
>> dtb_size = fdt_totalsize((void *)&__appended_dtb);
>>
>> /* last four bytes is always image size in little
>> endian */
>> - image_size = le32_to_cpup((void *)&__image_end - 4);
>> + image_size = get_unaligned_le32((void
>> *)&__image_end - 4);
>>
>> /* copy dtb to where the booted kernel will expect
>> it */
>> memcpy((void *)VMLINUX_LOAD_ADDRESS_ULL +
>> image_size,
>> --
>> 2.29.2
>>
>
>
> --
> Thanks,
> ~Nick Desaulniers


\
 
 \ /
  Last update: 2020-12-17 03:37    [W:0.103 / U:1.076 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site