question: i386 ld.gold reports "internal error" linking files with SHT_RELA section

question: i386 ld.gold reports "internal error" linking files with SHT_RELA section

[Xi Ruoyao]

Building SeaMonkey on i686 results:

    "/usr/bin/ld.gold: internal error in relocate_section, at ../../gold/i386.cc:3683"

i386.cc:3683 is:

    gold_assert(sh_type == elfcpp::SHT_REL);

AFAIK i386 normally use SHT_REL and x86_64 normally use SHT_RELA.  But
we've found that nasm-2.15.05 generates SHT_RELA for i386 as well.

So should SHT_RELA be allowed for i386 at all?  If "yes" we should fix
it for gold, if "no" we should fix it for nasm.
-- 
Xi Ruoyao <xry111@xry111.site>
School of Aerospace Science and Technology, Xidian University

Re: question: i386 ld.gold reports "internal error" linking files with SHT_RELA section

[H.J. Lu]

On Tue, Aug 30, 2022 at 7:39 AM Xi Ruoyao via Binutils
<binutils@sourceware.org> wrote:
>
> Building SeaMonkey on i686 results:
>
>     "/usr/bin/ld.gold: internal error in relocate_section, at ../../gold/i386.cc:3683"
>
> i386.cc:3683 is:
>
>     gold_assert(sh_type == elfcpp::SHT_REL);
>
> AFAIK i386 normally use SHT_REL and x86_64 normally use SHT_RELA.  But
> we've found that nasm-2.15.05 generates SHT_RELA for i386 as well.
>
> So should SHT_RELA be allowed for i386 at all?  If "yes" we should fix
> it for gold, if "no" we should fix it for nasm.
> --
> Xi Ruoyao <xry111@xry111.site>
> School of Aerospace Science and Technology, Xidian University

 SHT_RELA shouldn't be used for i386.

-- 
H.J.

Re: question: i386 ld.gold reports "internal error" linking files with SHT_RELA section

[Pierre Labastie]

On Tue, 2022-08-30 at 11:21 -0700, H.J. Lu wrote:
> On Tue, Aug 30, 2022 at 7:39 AM Xi Ruoyao via Binutils
> <binutils@sourceware.org> wrote:
> > 
> > Building SeaMonkey on i686 results:
> > 
> >     "/usr/bin/ld.gold: internal error in relocate_section, at
> > ../../gold/i386.cc:3683"
> > 
> > i386.cc:3683 is:
> > 
> >     gold_assert(sh_type == elfcpp::SHT_REL);
> > 
> > AFAIK i386 normally use SHT_REL and x86_64 normally use SHT_RELA.  But
> > we've found that nasm-2.15.05 generates SHT_RELA for i386 as well.
> > 
> > So should SHT_RELA be allowed for i386 at all?  If "yes" we should fix
> > it for gold, if "no" we should fix it for nasm.
> > --
> > Xi Ruoyao <xry111@xry111.site>
> > School of Aerospace Science and Technology, Xidian University
> 
>  SHT_RELA shouldn't be used for i386.
> 

But it is accepted by ld.bfd...

FWIIW, only the dwarf debug sections are generated as SHT_RELA by nasm.

Pierre Labastie

Re: question: i386 ld.gold reports "internal error" linking files with SHT_RELA section

[H.J. Lu]

On Tue, Aug 30, 2022 at 11:53 AM Pierre Labastie
<pierre.labastie@neuf.fr> wrote:
>
> On Tue, 2022-08-30 at 11:21 -0700, H.J. Lu wrote:
> > On Tue, Aug 30, 2022 at 7:39 AM Xi Ruoyao via Binutils
> > <binutils@sourceware.org> wrote:
> > >
> > > Building SeaMonkey on i686 results:
> > >
> > >     "/usr/bin/ld.gold: internal error in relocate_section, at
> > > ../../gold/i386.cc:3683"
> > >
> > > i386.cc:3683 is:
> > >
> > >     gold_assert(sh_type == elfcpp::SHT_REL);
> > >
> > > AFAIK i386 normally use SHT_REL and x86_64 normally use SHT_RELA.  But
> > > we've found that nasm-2.15.05 generates SHT_RELA for i386 as well.
> > >
> > > So should SHT_RELA be allowed for i386 at all?  If "yes" we should fix
> > > it for gold, if "no" we should fix it for nasm.
> > > --
> > > Xi Ruoyao <xry111@xry111.site>
> > > School of Aerospace Science and Technology, Xidian University
> >
> >  SHT_RELA shouldn't be used for i386.
> >
>
> But it is accepted by ld.bfd...

Do you have .o files for ld.bfd I can take a look at?

> FWIIW, only the dwarf debug sections are generated as SHT_RELA by nasm.
>
> Pierre Labastie



-- 
H.J.

Re: question: i386 ld.gold reports "internal error" linking files with SHT_RELA section

[Xi Ruoyao]

[-- Attachment #1: Type: text/plain, Size: 918 bytes --]

On Tue, 2022-08-30 at 13:36 -0700, H.J. Lu wrote:

> > >  SHT_RELA shouldn't be used for i386.
> > 
> > But it is accepted by ld.bfd...
> 
> Do you have .o files for ld.bfd I can take a look at?
> 
> > FWIIW, only the dwarf debug sections are generated as SHT_RELA by
> > nasm.

Hi,

$ cat t.asm
section .text
extern printf
global main
main:
  push ebp
  mov ebp, esp
  push msg
  call printf
  add esp, 4
  leave
  ret

section .data
  msg db 'Hello, world!', 13, 10, 0

section .note.GNU-stack noalloc noexec nowrite progbits
$ nasm t.asm -f elf32 -F dwarf
$ cc t.o && ./a.out 
Hello, world!
$ cc -fuse-ld=gold t.o 
/usr/bin/ld.gold: internal error in relocate_section, at ../../gold/i386.cc:3683
collect2: error: ld returned 1 exit status

I'll attach "t.asm' and "t.o.gz".

-- 
Xi Ruoyao <xry111@xry111.site>
School of Aerospace Science and Technology, Xidian University

[-- Attachment #2: t.o.gz --]
[-- Type: application/gzip, Size: 586 bytes --]

[-- Attachment #3: t.asm --]
[-- Type: text/plain, Size: 232 bytes --]

section .text
extern printf
global main
main:
  push ebp
  mov ebp, esp
  push msg
  call printf
  add esp, 4
  leave
  ret

section .data
  msg db 'Hello, world!', 13, 10, 0

section .note.GNU-stack noalloc noexec nowrite progbits

Re: question: i386 ld.gold reports "internal error" linking files with SHT_RELA section

[H.J. Lu]

On Tue, Aug 30, 2022 at 9:43 PM Xi Ruoyao <xry111@xry111.site> wrote:
>
> On Tue, 2022-08-30 at 13:36 -0700, H.J. Lu wrote:
>
> > > >  SHT_RELA shouldn't be used for i386.
> > >
> > > But it is accepted by ld.bfd...
> >
> > Do you have .o files for ld.bfd I can take a look at?
> >
> > > FWIIW, only the dwarf debug sections are generated as SHT_RELA by
> > > nasm.
>
> Hi,
>
> $ cat t.asm
> section .text
> extern printf
> global main
> main:
>   push ebp
>   mov ebp, esp
>   push msg
>   call printf
>   add esp, 4
>   leave
>   ret
>
> section .data
>   msg db 'Hello, world!', 13, 10, 0
>
> section .note.GNU-stack noalloc noexec nowrite progbits
> $ nasm t.asm -f elf32 -F dwarf
> $ cc t.o && ./a.out
> Hello, world!
> $ cc -fuse-ld=gold t.o
> /usr/bin/ld.gold: internal error in relocate_section, at ../../gold/i386.cc:3683
> collect2: error: ld returned 1 exit status
>
> I'll attach "t.asm' and "t.o.gz".
>
> --
> Xi Ruoyao <xry111@xry111.site>
> School of Aerospace Science and Technology, Xidian University

What happened are

1. nasm generates RELA relocation for debug sections.  But addend
in RELA relocation is 0:

Relocation section '.rela.debug_info' at offset 0x3a0 contains 5 entries:
 Offset     Info    Type                Sym. Value  Symbol's Name + Addend
00000006  00000501 R_386_32               00000000   .debug_abbrev + 0
0000000c  00000201 R_386_32               00000000   .text + 0
00000010  00000201 R_386_32               00000000   .text + 0
00000014  00000601 R_386_32               00000000   .debug_line + 0
0000002c  00000201 R_386_32               00000000   .text + 0

and addend is stored at the relocation offset like REL relocation:

Hex dump of section '.debug_info':
  0x00000000 31000000 02000000 00000401 00000000 1...............
  0x00000010 12000000 00000000 782e7300 4e41534d ........x.s.NASM
  0x00000020 20322e31 352e3035 00018002 00000000  2.15.05........
  0x00000030 00000000 00

2. ld.bfd treats RELA relocation like REL relocation.  It uses addend at
the relocation offset, not in RELA relocation.

It just works with ld.bfd by accident.  A linker, like gold, can disallow RELA
relocation.  A linker, like lld, can use addend in RELA relocation and addend
at the relocation offset is ignored.


-- 
H.J.

Re: question: i386 ld.gold reports "internal error" linking files with SHT_RELA section

[Xi Ruoyao]

On Wed, 2022-08-31 at 12:20 -0700, H.J. Lu wrote:

> It just works with ld.bfd by accident.  A linker, like gold, can disallow RELA
> relocation.  A linker, like lld, can use addend in RELA relocation and addend
> at the relocation offset is ignored.

Thanks, we've reported the issue to NASM:
https://bugzilla.nasm.us/show_bug.cgi?id=3392807

-- 
Xi Ruoyao <xry111@xry111.site>
School of Aerospace Science and Technology, Xidian University

Re: question: i386 ld.gold reports "internal error" linking files with SHT_RELA section

[H.J. Lu]

On Wed, Aug 31, 2022 at 5:47 PM Xi Ruoyao <xry111@xry111.site> wrote:
>
> On Wed, 2022-08-31 at 12:20 -0700, H.J. Lu wrote:
>
> > It just works with ld.bfd by accident.  A linker, like gold, can disallow RELA
> > relocation.  A linker, like lld, can use addend in RELA relocation and addend
> > at the relocation offset is ignored.
>
> Thanks, we've reported the issue to NASM:
> https://bugzilla.nasm.us/show_bug.cgi?id=3392807
>

I created a pull request:

https://github.com/netwide-assembler/nasm/pull/37

to fix this issue.

-- 
H.J.