Non-PIC call sequence in shared objects that call libmvec functions.

[H.J. Lu]

On Tue, Feb 9, 2016 at 8:20 AM, Carlos O'Donell <carlos@redhat.com> wrote:
> Igor,
>
> Shouldn't this be a PIC call sequence to support
> creating shared objects that call through the alias?
>
> diff --git a/sysdeps/x86_64/fpu/svml_finite_alias.S b/sysdeps/x86_64/fpu/svml_finite_alias.S
> index 0062fe4..61725b7 100644
> --- a/sysdeps/x86_64/fpu/svml_finite_alias.S
> +++ b/sysdeps/x86_64/fpu/svml_finite_alias.S
> @@ -23,7 +23,7 @@
>
>  #define ALIAS_IMPL(alias, target) \
>  ENTRY (alias); \
> -       call target; \
> +       jmp target@plt; \
>         ret; \
>  END (alias)
>
> With GCC6 generating vectorized calls now we're seeing package build
> failures because having shared libraries call these entry points fails
> with the following:
>
> /usr/bin/ld: /usr/lib64/libmvec_nonshared.a(svml_finite_alias.oS):
> relocation R_X86_64_PC32 against undefined symbol `_ZGVbN2v_log@@GLIBC_2.22'
> can not be used when making a shared object; recompile with -fPIC
>

Would you mind opening a glibc bug?

Thanks.


H.J.