Re: ld: spurious empty .plt/.rela.plt sections generated

[H.J. Lu]

On Sat, Apr 30, 2011 at 4:57 PM, Alan Modra <address@hidden> wrote:
> On Fri, Apr 29, 2011 at 05:07:33PM -0700, Roland McGrath wrote:
>> I am seeing a strange thing.  The (BFD) linker spuriously generates .plt
>> and .rela.plt sections (both empty) for a link that has no need for them.
>> It's reproduced with a trivial example, and seen on today's binutils trunk.
>> Gold does not produce the empty sections.
>
> HJ, please verify that this fix doesn't break anything else.
>
> Index: bfd/elf64-x86-64.c
> ===================================================================
> RCS file: /cvs/src/src/bfd/elf64-x86-64.c,v
> retrieving revision 1.223
> diff -u -p -r1.223 elf64-x86-64.c
> --- bfd/elf64-x86-64.c  23 Apr 2011 01:16:31 -0000      1.223
> +++ bfd/elf64-x86-64.c  30 Apr 2011 23:50:57 -0000
> @@ -1261,7 +1261,9 @@ elf_x86_64_check_relocs (bfd *abfd, stru
>            case R_X86_64_PLT32:
>            case R_X86_64_GOTPCREL:
>            case R_X86_64_GOTPCREL64:
> -             if (!_bfd_elf_create_ifunc_sections (abfd, info))
> +             if (htab->elf.dynobj == NULL)
> +               htab->elf.dynobj = abfd;
> +             if (!_bfd_elf_create_ifunc_sections (htab->elf.dynobj, info))
>                return FALSE;
>              break;
>            }
> Index: bfd/elf32-i386.c
> ===================================================================
> RCS file: /cvs/src/src/bfd/elf32-i386.c,v
> retrieving revision 1.251
> diff -u -p -r1.251 elf32-i386.c
> --- bfd/elf32-i386.c    23 Apr 2011 01:16:31 -0000      1.251
> +++ bfd/elf32-i386.c    30 Apr 2011 23:50:59 -0000
> @@ -1376,7 +1376,9 @@ elf_i386_check_relocs (bfd *abfd,
>            case R_386_PLT32:
>            case R_386_GOT32:
>            case R_386_GOTOFF:
> -             if (!_bfd_elf_create_ifunc_sections (abfd, info))
> +             if (htab->elf.dynobj == NULL)
> +               htab->elf.dynobj = abfd;
> +             if (!_bfd_elf_create_ifunc_sections (htab->elf.dynobj, info))
>                return FALSE;
>              break;
>            }

It works.

Thanks.


-- 
H.J.