Re: [PATCH] Add support for new ABI choice e_flags values on ARM

[Steve McIntyre <steve dot mcintyre at linaro dot org>]

On Tue, Sep 04, 2012 at 06:25:55PM +0100, Steve McIntyre wrote:
>On Thu, Aug 30, 2012 at 01:21:42PM +0000, Joseph S. Myers wrote:
>>On Thu, 30 Aug 2012, Steve McIntyre wrote:
>>
>>> gas:
>>>     * config/tc-arm.c (md_begin): Use the new ARM
>>>     hard-float/soft-float ABI flags for EABI_VER5
>>
>>This seems wrong; the ABI is that these flags are only set in ET_EXEC and 
>>ET_DYN objects.  This means the assembler should have nothing to do with 
>>them; rather, the linker should set them based on the build attributes in 
>>its input files, if producing ET_EXEC or ET_DYN output.  (And there should 
>>be testcases added to the linker testsuite for this: that they are set 
>>for such output, but not for ld -r links.)
>
>Hi,
>
>Updated patch attached including test cases for ld. They work for me
>OK (explicitly testing both positive and negative cases for the ELF
>type and the ABI version), but it's the first time I've worked on
>input for dejagnu so review here would be great! My test cases work
>ok, but I'm seeing a lot of other failures unrelated to my changes; I
>assume that's expected.
>
>I've also tried to get gold to do the right thing here, hooking into
>Target_arm<big_endian>::do_adjust_elf_header(). But I don't see how
>that's ever called; my testing shows it's clearly not. I'm also a
>little baffled as to why gold's testsuite is so utterly different from
>the one for ld - can anyone explain that please?

Ping?

Cheers,
-- 
Steve McIntyre                                steve.mcintyre@linaro.org
<http://www.linaro.org/> Linaro.org | Open source software for ARM SoCs