Re: 3.2 PATCH: Support O32 ABI on IRIX 6

[Daniel Jacobowitz <drow at mvista dot com>]

On Wed, Jun 19, 2002 at 03:19:10PM +0200, Rainer Orth wrote:
> Eric Christopher writes:
> 
> > The patches themselves look fine, as long as we're thinking that we'll
> > still need to have a separate compiler for the o32 bits. I'd like the
> > CPP_PREDEFINES/SUBTARGET_CPP_SPEC etc, stuff removed, but in general
> 
> I've already done so, but still need to re-bootstrap with this change.  I
> got distracted by various other bootstrap failures on other platforms,
> though.
> 
> > they are fine. I was wondering more on your thoughts as to this approach
> > as opposed to attempting to support o32/n32/n64 on irix6. I believe more
> 
> I've started with the current approach of using a separate o32-only
> configuration since I can get this ready now.  I'm not sure I'll be able to
> complete a common three-ABI IRIX 6 configuration in time for the 3.2
> release, so I'd like to get in what I have now.
> 
> > of the mechanisms may be there now than there were in the past and this
> > should make it easier.
> 
> Probably: I hope to start looking into this soon.
> 
> > > This section is created by mips.c (mips_asm_file_start) [TARGET_GAS &&
> > > OBJECT_FORMAT_ELF].  Contrary to the comment there, current gdb or bfd
> > > don't seem to need or use this information, so it may well be possible to
> > > get rid of this code completely, unless there are other known uses.
> > > 
> > 
> > I added this code when I was adding the mips32/mips64 compiler bits in
> > as a separate way of determining ABI since we had/were running out of
> > bits. It was due to have separate code in gdb that I should _really_
> > resurrect, but wasn't multi-arched at the time. Kind of an odd warning
> 
> Mainline gdb now has code to deal with those sections:
> 
> 	http://sources.redhat.com/ml/gdb-patches/2002-06/msg00135.html
> 
> > though since I don't get that normally. We could probably conditionalize
> 
> I seem to get this warning from the O32 ld only.
> 
> > the code as being !TARGET_SGI though.
> 
> Right: the three ABIs in use on that platform should be easily recognizable
> on their own, without special .mdebug.<ABI> sections.  Thus I'll go for
> this approach in a subsequent patch.

Do all three have explicit ABI flags in the ELF header?  I don't know
how SGI handles that.  Otherwise, please don't remove the
.mdebug.<ABI>. sections....


-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer