This is the mail archive of the
mailing list for the binutils project.
Re: Setting GNU OSABI on ARM
- From: Will Newton <will dot newton at linaro dot org>
- To: nick clifton <nickc at redhat dot com>
- Cc: "binutils at sourceware dot org" <binutils at sourceware dot org>
- Date: Tue, 30 Apr 2013 14:43:47 +0100
- Subject: Re: Setting GNU OSABI on ARM
- References: <CANu=DmgXKWu+7JzmSRbdmwLApiKENYc+CP2GV-9p5jcwhN311A at mail dot gmail dot com> <517FC1E4 dot 8010207 at redhat dot com>
On 30 April 2013 14:06, nick clifton <firstname.lastname@example.org> wrote:
>> I was wondering if someone could help me figure out why all
>> architectures except ARM seem to set GNU OSABI when GNU indirect
>> function symbols are present?
>> The below patch sets GNU OSABI on ARM and lets us run the generic
>> ifunc tests (that run on i386, x86_64, sparc and powerpc). Is there
>> some historical reason why this is a bad idea?
> I do not know of any historical reason for this. I suspect that it was just
> an accident of the development of ifunc support.
> There may be a practical issue however, in that changing the EI_OSABI field
> may break OSes that expect it to be set to ELOSABI_ARM So whilst
> technically I have no objections to the patch I think that it would be wise
> to provide a configure time option to restore the original behavior.
This was my concern, whether or not anything relies on the current
OSABI behaviour. I am not sure if anything other than Linux actually
uses STT_GNU_IFUNC either.
I'm currently implementing indirect function support for AArch64 so I
would be interested if there was any rationale to the decision, as
otherwise I will just make it use the GNU OSABI. ;-)
Toolchain Working Group, Linaro