This is the mail archive of the
binutils@sources.redhat.com
mailing list for the binutils project.
Re: selective[1245] xfail on x86-64
On Thu, May 30, 2002 at 06:55:32PM +0200, Andreas Schwab wrote:
> Alan Modra <amodra@bigpond.net.au> writes:
>
> |> On Thu, May 30, 2002 at 02:09:54PM +0100, Nick Clifton wrote:
> |> > Hi Alan,
> |> >
> |> > > > Hmm - well at the very least I think that you ought to add a comment
> |> > > > as well, explaining that it is relocs in the .eh_frame section that
> |> > > > are keeping the redundant symbols in the executables. It would also
> |> > > > be good to add a FIXME note to the x86_64 sources somewhere saying
> |> > > > that this ought to be corrected some day.
> |> > >
> |> > > It's not specific to x86_64. I see the same thing on powerpc64, and
> |> > > I think other targets will suffer the same problem.
> |> >
> |> > Ok - well then perhaps there ought to be a more generic test in
> |> > selective.exp. Something like:
> |>
> |> Or maybe we could just treat .eh_frame specially in elf_gc_mark.
> |> Andreas, does this cure the x86_64 problem?
>
> Doesn't work:
>
> tmpdir/1.o(.eh_frame+0x38): undefined reference to `local symbols in discarded section .text.dropme1'
Yep. Alan, if you want this to work you'll need to either:
- ignore discarded locals if you're in .eh_frame
- properly edit .eh_frame; I guess we only do this for .debug_frame.
>
> |> (It doesn't work
> |> on powerpc64 because the .opd section refers to all functions.
>
> Can't this be fixed by splitting .opd in the same way as .text?
I'd think it would be easier to remove entries from .opd automatically
when we GC a function.
--
Daniel Jacobowitz Carnegie Mellon University
MontaVista Software Debian GNU/Linux Developer