[PATCH] backends: aarch64 always has _GLOBAL_OFFSET_TABLE_ point to .got[0].

Mark Wielaard mjw at redhat.com
Fri Dec 20 15:33:33 UTC 2013


Hi,

On Thu, 2013-12-19 at 16:17 +0100, Mark Wielaard wrote:
> Like some other architectures aarch64 has a special rule for the
> _GLOBAL_OFFSET_TABLE_ symbol. Even if there is a .plt.got section the symbol
> value still points to the start of the .got section. This is also what the
> dynamic linker expects.
> 
> See https://sourceware.org/ml/libc-ports/2013-06/msg00057.html

> +2013-12-19  Mark Wielaard  <mjw at redhat.com>
> +
> +	* aarch64_init.c (aarch64_init): Hook check_special_symbol.
> +	* aarch64_symbol.c (aarch64_check_special_symbol): New function.

I pushed this to master, with one typo fixed, see below.
It worked nicely on the fedora aarch64 build:
http://arm.koji.fedoraproject.org/koji/buildinfo?buildID=182251
build logs show all tests passing.

> +/* If this is the _GLOBAL_OFFSET_TABLE_ symbol, then it should point to
> +   .got[0] even if there is a .got.plt section.  */
> +bool
> +aarch64_check_special_symbol (Elf *elf, GElf_Ehdr *ehdr, const GElf_Sym *sym,
> +                              const char *name, const GElf_Shdr *destshdr)
> +{
> +  if (name != NULL
> +      && strcmp (name, "_GLOBAL_OFFSET_TABLE_") == 0)
> +    {
> +      const char *sname = elf_strptr (elf, ehdr->e_shstrndx, destshdr->sh_name);
> +      if (sname != NULL && strcmp (sname, ".got.plt") == 0)
> +	{
> +	  Elf_Scn *scn = NULL;
> +	  while ((scn = elf_nextscn (elf, scn)) != NULL)
> +	    {
> +	      GElf_Shdr shdr_mem;
> +	      GElf_Shdr *shdr = gelf_getshdr (scn, &shdr_mem);
> +	      sname = elf_strptr (elf, ehdr->e_shstrndx, shdr->sh_name);
> +	      if (name != NULL && strcmp (sname, ".got") == 0)
                    ^
                    That should be sname of course.

> +		return sym->st_value == shdr->sh_addr;
> +	    }
> +	}
> +    }
> +
> +  return false;
> +}




More information about the elfutils-devel mailing list