[PATCH] X86: Work around a bfd ld quirk wrt GOTTPOFF relocation processing

[H.J. Lu]

On Mon, Jan 5, 2015 at 7:14 AM, Rafael Espíndola
<rafael.espindola at gmail.com> wrote:
>> It is due to the requirement of the REX prefix for IE->LE optimization:
>
> Not sure I follow.
>
> Given
>
>
> main:
>         movl    %fs:0, %eax
>         addl    external_ie at GOTTPOFF(%rip), %eax
>         retq
>
> gold produces
>
> 0000000000400590 <main>:
>   400590: 64 8b 04 25 00 00 00 mov    %fs:0x0,%eax
>   400597: 00
>   400598: 8d 80 fc ff ff ff     lea    -0x4(%rax),%eax
>   40059e: c3                   retq
>
> What is wrong with this transformation? Why can't bfd ld do the same?

See:

https://sourceware.org/bugzilla/show_bug.cgi?id=17795

>
> Adding a dummy rex also doesn't help. BFD errors on
>
>
> main:
>         movq    %fs:0, %rax
>         rex addl        external_ie at GOTTPOFF(%rip), %eax
>         retq

This is supported only for x32 since x32 is limited to 32-bit
address space.   How do you know R_X86_64_GOTTPOFF
won't overflow 32-bit register in 64-bit address space?

>
> In any case. This is at least a bfd ld limitation and we should work
> around it. This patch LGTM.
>


-- 
H.J.