Re: GCC -pie patch (kinda) ....

["Joseph S. Myers" <joseph@codesourcery.com>]

On Sun, 2 Mar 2008, Michael Schmitz wrote:

> > > elf-init.c:(.text+0x6e): undefined reference to `__fini_array_end'
> > > elf-init.c:(.text+0x76): undefined reference to `__fini_array_start'
> > > elf-init.c:(.text+0x8a): undefined reference to `__fini_array_start'
> > > collect2: ld returned 1 exit status
> >
> > These errors generally (on whatever ELF platform you encounter them)
> > indicate a linker script problem, since these symbols are meant to be
> > defined by a linker script.  In this case, a missing linker script, patch:
> > <http://sourceware.org/ml/binutils/2008-03/msg00006.html>.
> 
> Thanks for tracking this down!
> 
> Should not distract us from TLS, though. The plan was to backport TLS
> support from the coldfire port, right? Is there anything to look at, yet?

As I noted <http://lists.debian.org/debian-68k/2008/02/msg00212.html>, 
we're finalizing the kernel interface with Freescale's kernel engineers, 
who will be implementing the kernel side of things.  We know from 
experience (having previously done the TLS and NPTL implementations for 
ARM and MIPS) that changes to one part of the implementation, and to the 
specification, may turn out to be needed in the course of another part of 
the implementation, and that kernel bugs are likely to show up in 
validation, so it doesn't make sense to try to contribute patches for one 
component until we have everything working properly together.  After that, 
we expect to start posting patches in the order binutils, GCC, EGLIBC (the 
kernel patches are up to Freescale).  (Finding problematic interations in 
the course of implementation is especially likely when breaking new 
ground: m68k is the first target using a vDSO for atomic and thread 
operations, with possible complications when such operations are used 
early in ld.so startup.)

-- 
Joseph S. Myers
joseph@codesourcery.com