Re: Relocs against unalloced sections in shared libraries

[Daniel Jacobowitz <drow at mvista dot com>]

On Thu, Jan 09, 2003 at 04:39:08PM -0800, Geoff Keating wrote:
> > From: Andreas Schwab <schwab@suse.de>
> > Date: 10 Jan 2003 01:29:50 +0100
> 
> > Geoff Keating <geoffk@geoffk.org> writes:
> > 
> > > Isn't this necessary for proper handling of DWARF2 debug information?
> > 
> > The debugger will know which bytes to relocate by looking at the
> > structure while reading the debug information from a shared library.
> 
> Hmmm.  Have you asked the GDB people about this?  I believe the
> conclusion was that it kind-of does this now, and it kind-of works,
> but in the long run it should be looking at the relocations.

Yes.  I don't know if it's possible to reproduce the situation in a
shared library on PPC (I haven't tried) but there are ET_REL objects
on PPC that _can not_ be debugged without properly obeying the
relocation information.  That's because there can be relocations
against variables instead of against section symbols.

The GDB code to do this has been finished for some time and is still
waiting for review.  I'll nag them about it again in a couple of weeks.

> > No other target emits such relocations.  Currently, the ppc linker

Richard vehemently maintains that they should.  He's specifically
vetoed this change in the past.

> > will set DT_TEXTREL if there are any DWARF2 debugging sections in a
> > shared library, which is just wrong.
> 
> Yes, the linker should ignore relocs against unallocated sections.
> I thought it already did, we have had this entire discussion several
> times before.

Yes, that's a different problem.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer