[PATCH] ld: provide __ehdr_start magic symbol

[Roland McGrath]

On Tue, Jun 26, 2012 at 6:19 PM, Alan Modra <amodra@gmail.com> wrote:
> On Tue, Jun 26, 2012 at 06:21:43AM +0200, Hans-Peter Nilsson wrote:
> > cris-axis-linux-gnu.  Instead only targets expected to hold
> > phdrs in the address space should be included.  Other such tests
> > use "#target: *-*-linux* *-*-gnu*", which seems appropriate here
> > too.

I had thought that the original canonical ELF layouts with headers in text
were the norm and there would be only a few outliers.  I guess there are
many.  It's certainly far from being only *-*-linux* and *-*-gnu* that have this
layout, though.

> gas/testsuite/
>        * gas/i386/rep-ret.s: Zero pad section.
>        * gas/i386/rep-ret.d: Update.
> ld/testsuite/
>        * ld-elf/ehdr_start.s: Use data rather than rodata.
>        * ld-elf/ehdr_start.d: Run on linux and gnu targets only.

I added *-*-nacl* too, since that's the target whose needs motivated
the feature.

> The above fixes the failures, except for the frv-linux one, which now
> fails with
>
> ./ld-new: BFD (GNU Binutils) 2.22.52.20120627 assertion fail /src/binutils-current/bfd/elf32-frv.c:1333
> ./ld-new: BFD (GNU Binutils) 2.22.52.20120627 assertion fail /src/binutils-current/bfd/elf32-frv.c:1324
> LINKER BUG: .rofixup section size mismatch
> ./ld-new: final link failed: Nonrepresentable section on output

No idea what that might be about.  The test is quite simple.


Thanks,
Roland