Re: ix86 targets without BFD64

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Mon, Oct 23, 2017 at 5:11 AM, Jan Beulich <JBeulich@suse.com> wrote:
>>>> On 23.10.17 at 13:13, <hjl.tools@gmail.com> wrote:
>> On Sun, Oct 22, 2017 at 11:53 PM, Jan Beulich <JBeulich@suse.com> wrote:
>>> Hello,
>>>
>>> there being quite a few BFD64 conditionals in tc-i386.c, I would
>>> conclude that things are supposed to work fine when doing a
>>> purely 32-bit build (which happen to result when I cross build on
>>> cygwin, but of course by not forcing --enable-64-bit-bfd I can
>>> also see this on a native Linux build on a 32-bit distro). Yet each of
>>> the last two lines of
>>>
>>>         .text
>>> _start:
>>>         mov     0x87654321, %eax
>>>         mov     $0x87654321, %eax
>>>
>>>         .code64
>>>         movabs  0x87654321, %al
>>>         movabs  $0x87654321, %rax
>>>
>>> trigger the abort() in offset_in_range(). Since the chosen target
>>> automatically enables elf32_x86_64, the solution can't possibly be
>>> to disallow .code64 in such a build. All other alternatives I can
>>> think of would, however, appear to require a full audit of all 64-bit
>>> immediate / displacement processing in order to make sure all valid
>>> code (i.e. anything not requiring 64-bit relocations) is accepted.
>>>
>>> Thoughts?
>>
>> We can disable elf32_x86_64 and disallow .code64 if bfd_vma is 32 bit.
>
> Hmm, interesting. Then why was it added there in the first place?
>

Since my i386 binutils is built with 64-bit BFD, it isn't a problem.
BTW, I don't believe elf32_x86_64 linker works correctly with 32-bit
BFD.


-- 
H.J.