Re: Linking very large programs with inadequate branch displacements

[Bill Yoder <byoder at cs dot utexas dot edu>]

Dear Alan, H.J., Nick, and H-P,

Thanks for your prompt help:

> Please check out elfNN_ia64_relax_section in elfxx-ia64.c.
> H.J.
>
> Also have a look at bfd/elf32- 
xstormy16.c:xstormy16_elf_check_relocs().
> Nick
>
> See elf32-hppa.c and elf64-ppc.c for one method of doing this, and
> elf32-ppc.c for another.
> Alan Modra
>
> Usually referred to as a "trampoline" or "stub".  For yet
> another implementation, see bfd/elf64-mmix.c, handling of
> R_MMIX_PUSHJ_STUBBABLE.  See also gas/config/tc-mmix.c for when
> these relocs are generated.
> brgds, H-P

I've looked at these examples, cooked an x86/Linux version of  
binutils for --target=ppc-elf32-linux, and run a number of gcc 3.2  
PPC-compiled files through the debuggable assembler and linker.

Alan, I can't figure out how to hit a breakpoint at  
ppc_elf_relax_section:elf32-ppc.c.  Is there a command line switch I  
should throw, or would you have some C/ASM source that will trigger  
this routine?

Bill

P.S. I appreciate the humor sprinkled in the sources:
"Oh dear, we can't reach a trampoline...", "If the branch and target  
are in the same section, you've got one honking big section...", etc.


>