Re: fix powerpc-eabisim -mrelocatable problem

[Geoff Keating <geoffk at geoffk dot org>]

> From: Franz Sirl <Franz.Sirl-kernel@lauterbach.com>
> Date: Mon, 5 Nov 2001 21:19:09 +0100
> Cc: pedwards@disaster.jaj.com
> 
> On Monday 05 November 2001 21:03, Geoffrey Keating wrote:
> > ...
> >
> > > /tmp/ccOGdRdk.s:4293: Error: Relocation cannot be done when using
> > > -mrelocatable
> >
> > ...
> >
> > This patch fixes this.  The problem was that we were putting
> > CONSTRUCTOR trees in .rodata even when they shouldn't have been; the
> > assembler catches this because it sees a relocation in a read-only
> > section that shouldn't be there.  (This has implications for -fpic on
> > linux systems too, but it doesn't cause a hard error, just a
> > performance penalty.)
> >
> > Phil, I suspect you weren't seeing it because you had the wrong
> > binutils version or something.
> >
> > Tests running on powerpc-eabisim, I'll commit when they complete.
> 
> Hmm, isn't this basically the same patch I suggested long ago? At that time 
> it was turned down because the testcase was C-only and was using a GNU 
> extension that is now part of C99. Thus the plan was to fix this in generic 
> code while completing C99 features in gcc3. What has changed now?

powerpc-eabi was failing to bootstrap, and other ports (well, Alpha)
seemed to have similar code.  I don't remember the previous
discussion.

-- 
- Geoffrey Keating <geoffk@geoffk.org> <geoffk@redhat.com>