Date:   Wed, 19 Aug 2020 20:23:10 -0700
From:   Yonghong Song <yhs@...com>
To:     Fāng-ruì Sòng <maskray@...gle.com>,
        Jiri Olsa <jolsa@...nel.org>
CC:     Nick Desaulniers <ndesaulniers@...gle.com>,
        Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Mark Wielaard <mjw@...hat.com>,
        Nick Clifton <nickc@...hat.com>,
        Jesper Dangaard Brouer <brouer@...hat.com>,
        Network Development <netdev@...r.kernel.org>,
        bpf <bpf@...r.kernel.org>, Martin KaFai Lau <kafai@...com>,
        Song Liu <songliubraving@...com>,
        Andrii Nakryiko <andriin@...com>,
        John Fastabend <john.fastabend@...il.com>,
        KP Singh <kpsingh@...omium.org>,
        clang-built-linux <clang-built-linux@...glegroups.com>
Subject: Re: [PATCH bpf-next] tools/resolve_btfids: Fix sections with wrong
 alignment



On 8/19/20 7:27 PM, Fāng-ruì Sòng wrote:
>>>>     section(36) .comment, size 44, link 0, flags 30, type=1
>>>>     section(37) .debug_aranges, size 45684, link 0, flags 800, type=1
>>>>      - fixing wrong alignment sh_addralign 16, expected 8
>>>>     section(38) .debug_info, size 129104957, link 0, flags 800, type=1
>>>>      - fixing wrong alignment sh_addralign 1, expected 8
>>>>     section(39) .debug_abbrev, size 1152583, link 0, flags 800, type=1
>>>>      - fixing wrong alignment sh_addralign 1, expected 8
>>>>     section(40) .debug_line, size 7374522, link 0, flags 800, type=1
>>>>      - fixing wrong alignment sh_addralign 1, expected 8
>>>>     section(41) .debug_frame, size 702463, link 0, flags 800, type=1
>>>>     section(42) .debug_str, size 1017571, link 0, flags 830, type=1
>>>>      - fixing wrong alignment sh_addralign 1, expected 8
>>>>     section(43) .debug_loc, size 3019453, link 0, flags 800, type=1
>>>>      - fixing wrong alignment sh_addralign 1, expected 8
>>>>     section(44) .debug_ranges, size 1744583, link 0, flags 800, type=1
>>>>      - fixing wrong alignment sh_addralign 16, expected 8
>>>>     section(45) .symtab, size 2955888, link 46, flags 0, type=2
>>>>     section(46) .strtab, size 2613072, link 0, flags 0, type=3
> 
> I think this is resolve_btfids's bug. GNU ld and LLD are innocent.
> These .debug_* sections work fine if their sh_addralign is 1.
> When the section flag SHF_COMPRESSED is set, the meaningful alignment
> is Elf64_Chdr::ch_addralign, after the header is uncompressed.
> 
> On Wed, Aug 19, 2020 at 2:30 PM Yonghong Song <yhs@...com> wrote:
>>
>>
>>
>> On 8/19/20 11:16 AM, Nick Desaulniers wrote:
>>> On Wed, Aug 19, 2020 at 10:36 AM Jiri Olsa <jolsa@...hat.com> wrote:
>>>>
>>>> On Wed, Aug 19, 2020 at 08:31:51AM -0700, Yonghong Song wrote:
>>>>>
>>>>>
>>>>> On 8/19/20 2:23 AM, Jiri Olsa wrote:
>>>>>> The data of compressed section should be aligned to 4
>>>>>> (for 32bit) or 8 (for 64 bit) bytes.
>>>>>>
>>>>>> The binutils ld sets sh_addralign to 1, which makes libelf
>>>>>> fail with misaligned section error during the update as
>>>>>> reported by Jesper:
>>>>>>
>>>>>>       FAILED elf_update(WRITE): invalid section alignment

Jiri,

Since Fangrui mentioned this is not a ld/lld bug, then changing
alighment from 1 to 4 might have some adverse effect for the binary,
I guess.

Do you think we could skip these .debug_* sections somehow in elf 
parsing in resolve_btfids? resolve_btfids does not need to read
these sections. This way, no need to change their alignment either.

Yonghong

>>>>>>
>>>>>> While waiting for ld fix, we can fix compressed sections
>>>>>> sh_addralign value manually.
>>>
>>> Is there a bug filed against GNU ld? Link?
>>>
>>>>>>
>>>>>> Adding warning in -vv mode when the fix is triggered:
>>>>>>
>>>>>>      $ ./tools/bpf/resolve_btfids/resolve_btfids -vv vmlinux
>>>>>>      ...
>>>>>>      section(36) .comment, size 44, link 0, flags 30, type=1
>>>>>>      section(37) .debug_aranges, size 45684, link 0, flags 800, type=1
>>>>>>       - fixing wrong alignment sh_addralign 16, expected 8
>>>>>>      section(38) .debug_info, size 129104957, link 0, flags 800, type=1
>>>>>>       - fixing wrong alignment sh_addralign 1, expected 8
>>>>>>      section(39) .debug_abbrev, size 1152583, link 0, flags 800, type=1
>>>>>>       - fixing wrong alignment sh_addralign 1, expected 8
>>>>>>      section(40) .debug_line, size 7374522, link 0, flags 800, type=1
>>>>>>       - fixing wrong alignment sh_addralign 1, expected 8
>>>>>>      section(41) .debug_frame, size 702463, link 0, flags 800, type=1
>>>>>>      section(42) .debug_str, size 1017571, link 0, flags 830, type=1
>>>>>>       - fixing wrong alignment sh_addralign 1, expected 8
>>>>>>      section(43) .debug_loc, size 3019453, link 0, flags 800, type=1
>>>>>>       - fixing wrong alignment sh_addralign 1, expected 8
>>>>>>      section(44) .debug_ranges, size 1744583, link 0, flags 800, type=1
>>>>>>       - fixing wrong alignment sh_addralign 16, expected 8
>>>>>>      section(45) .symtab, size 2955888, link 46, flags 0, type=2
>>>>>>      section(46) .strtab, size 2613072, link 0, flags 0, type=3
>>>>>>      ...
>>>>>>      update ok for vmlinux
>>>>>>
>>>>>> Another workaround is to disable compressed debug info data
>>>>>> CONFIG_DEBUG_INFO_COMPRESSED kernel option.
>>>>>
>>>>> So CONFIG_DEBUG_INFO_COMPRESSED is required to reproduce the bug, right?
>>>>
>>>> correct
>>>>
>>>>>
>>>>> I turned on CONFIG_DEBUG_INFO_COMPRESSED in my config and got a bunch of
>>>>> build failures.
>>>>>
>>>>> ld: drivers/crypto/virtio/virtio_crypto_algs.o: unable to initialize
>>>>> decompress status for section .debug_info
>>>>> ld: drivers/crypto/virtio/virtio_crypto_algs.o: unable to initialize
>>>>> decompress status for section .debug_info
>>>>> ld: drivers/crypto/virtio/virtio_crypto_algs.o: unable to initialize
>>>>> decompress status for section .debug_info
>>>>> ld: drivers/crypto/virtio/virtio_crypto_algs.o: unable to initialize
>>>>> decompress status for section .debug_info
>>>>> drivers/crypto/virtio/virtio_crypto_algs.o: file not recognized: File format
>>>>> not recognized
>>>>>
>>>>> ld: net/llc/llc_core.o: unable to initialize decompress status for section
>>>>> .debug_info
>>>>> ld: net/llc/llc_core.o: unable to initialize decompress status for section
>>>>> .debug_info
>>>>> ld: net/llc/llc_core.o: unable to initialize decompress status for section
>>>>> .debug_info
>>>>> ld: net/llc/llc_core.o: unable to initialize decompress status for section
>>>>> .debug_info
>>>>> net/llc/llc_core.o: file not recognized: File format not recognized
>>>>>
>>>>> ...
>>>>>
>>>>> The 'ld' in my system:
>>>>>
>>>>> $ ld -V
>>>>> GNU ld version 2.30-74.el8
>>>>>     Supported emulations:
>>>>>      elf_x86_64
>>>>>      elf32_x86_64
>>>>>      elf_i386
>>>>>      elf_iamcu
>>>>>      i386linux
>>>>>      elf_l1om
>>>>>      elf_k1om
>>>>>      i386pep
>>>>>      i386pe
>>>
>>> According to Documentation/process/changes.rst, the minimum supported
>>> version of GNU binutils for the kernels is 2.23.  Can you upgrade to
>>> that and confirm that you still observe the issue?  I don't want to
>>> spend time chasing bugs in old, unsupported versions of GNU binutils,
>>> especially as Jiri notes, 2.26 is required for
>>> CONFIG_DEBUG_INFO_COMPRESSED.  We can always strengthen the Kconfig
>>> check for it.  Otherwise, I'm not familiar with the observed error
>>> message.
>>
>> I built a "ld" with latest binutils-gdb repo and I can reproduced
>> the issue. Indeed applying the patch here fixed the issue. So
>> I think there is no need to investigate since upstream exhibits
>> the exact issue described here.
>>
>>>
>>>>> $
>>>>>
>>>>> Do you know what is the issue here?
>>>>
>>>> mine's: GNU ld version 2.32-31.fc31
>>>>
>>>> there's version info in commit:
>>>>     10e68b02c861 Makefile: support compressed debug info
>>>>
>>>>     Compress the debug information using zlib.  Requires GCC 5.0+ or Clang
>>>>     5.0+, binutils 2.26+, and zlib.
>>>>
>>>> cc-ing Nick Desaulniers, author of that patch.. any idea about the error above?
>>>>
>>>> thanks,
>>>> jirka
>>>>
>>>
>>>
> 
> 
> 

Powered by blists - more mailing lists