Re: [PATCH] Fix distinction of 32/64bit addresses in MIPS gas

[cgd at broadcom dot com]

mrg@cygnus.com ("matthew green") writes:
> sure, there are bugs, but there's no ignoring the fact that these
> tools have been used by a whole lot of people for a number of years
> quite successfullly.  it's simply true.

I don't believe the current versions will, even disregarding
unintentional issues which have cropped up over the past couple of
weeks or months.

I'm using a version of binutils a couple of months old to generate
(dynamically-linked) NetBSD/mips binaries, but alas there is an
intentional change in the ABI of the binaries generated by binutils,
which breaks compatibility with existing NetBSD/mips ELF binaries and
the existing NetBSD/mips shared loader (which, AFAICT, would have
worked with minimal configury tweaking, previously).


To go back through my notes, it was caused by one or the combination
of:

2000-10-13  Ulf Carlsson  <ulfc@engr.sgi.com>

        From  Ralf Baechle  <ralf@gnu.org>
        * elf32-mips.c (mips_elf_create_dynamic_relocation): New argument
        local_p.  Add symbol value only for non-R_MIPS_REL32 relocations
        against local symbols.
        (_bfd_mips_elf_finish_dynamic_sections): Undo patch from 2000-10-01.

2001-01-18  Maciej W. Rozycki  <macro@ds2.pg.gda.pl>

        * elf32-mips.c (mips_elf_create_dynamic_relocation): Undo patch
        from 2000-10-13.  Do not add the symbol's value for R_MIPS_REL32
        relocations against dynsym symbols.

but it's been a couple of months since I looked into it, and I don't
have all of my notes and source trees handy to tell which it was.  (I
pointed it out to a few NetBSD folks back in early July.)



cgd