See

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14660

The problem is

(gdb) bt
#0  0xb77e9942 in _dl_sysinfo_int80 () at rtld.c:571
#1  0xb76e9165 in *__GI_raise (sig=6)
    at ../nptl/sysdeps/unix/sysv/linux/raise.c:67
#2  0xb76ea949 in *__GI_abort () at ../sysdeps/generic/abort.c:88
#3  0x08049dbf in _Unwind_SetGR (context=0x0, index=6, val=0)
    at /net/gnu/export/gnu/src/gcc-3.4/gcc/gcc/unwind-dw2.c:240
#4  0x08049aa2 in __gcc_personality_v0 (version=1, actions=10,
    exception_class=3221213696, ue_header=0x0, context=0xbfffd020)
    at /net/gnu/export/gnu/src/gcc-3.4/gcc/gcc/unwind-c.c:190
#5  0xb76ae7f6 in _Unwind_ForcedUnwind_Phase2 (exc=0xb77d7b40,
    context=0xbfffd020) at unwind.inc:180
#6  0xb76ae8d7 in _Unwind_ForcedUnwind (exc=0xb77d7b40, stop=0,
    stop_argument=0x0) at unwind.inc:211
#7  0xb77e3ab2 in _Unwind_ForcedUnwind (exc=0xb77d7b40, stop=0xb77d7b40,
    stop_argument=0xb77d7b40)
    at ../nptl/sysdeps/pthread/unwind-forcedunwind.c:88
#8  0xb77e18f1 in *__GI___pthread_unwind (buf=0xb77d7b40) at unwind.c:130
#9  0xb77e044d in __pthread_setcanceltype (type=1, oldtype=0x0)
    at pthreadP.h:149
#10 0x08049012 in do_test () at tst-cleanup0.c:62
#11 0x08049354 in main (argc=0, argv=0xbfffd274) at test-skeleton.c:268
(gdb) info shared
From        To          Syms Read   Shared Object Library
0xb77dc490  0xb77e3f60  Yes        
/export/build/gnu/glibc-nptl-3.4/build-i686-linux/nptl/libpthread.so.0
0xb76d60a0  0xb77b09ac  Yes        
/export/build/gnu/glibc-nptl-3.4/build-i686-linux/libc.so.6
0xb77e9940  0xb77fbd6f  Yes        
/export/build/gnu/glibc-nptl-3.4/build-i686-linux/elf/ld-linux.so.2
0xb76ab690  0xb76b0ab0  Yes         /lib/libgcc_s.so.1

As you can see, _Unwind_ForcedUnwind comes from libgcc_s.so.1 and
_Unwind_SetGR comes from executable. They won't work together since
they use their own copy of dwarf_reg_size_table. The relevant gcc change
is

http://gcc.gnu.org/ml/gcc-patches/2003-10/msg00013.html

------- Additional Comment #1 From H.J. Lu 2004-04-02 16:12 -------

Created an attachment (id=44)
A patch to use --as-needed

This patch uses --as-needed when it is available. I tested it on ia32 and
ia64. It works fine for me.

------- Additional Comment #2 From Ulrich Drepper 2004-04-03 08:54 -------

The patch does too much.  There is no need for HAVE_AS_NEEDED.  I've removed
those parts and applied the rest.

------- Additional Comment #3 From H.J. Lu 2004-04-09 16:39 -------

There are a few typoes in my previous patch. It should check if -lgcc_s exists
before using it. Also it should use -Wl,--as-needed, instead of just --as-needed.

------- Additional Comment #4 From H.J. Lu 2004-04-09 16:40 -------

Created an attachment (id=50)
A patch to check -lgcc_s and add -Wl,

------- Additional Comment #5 From Nigel Kukard 2004-04-28 17:32 -------

any updates on getting glibc 2.3.2 working under gcc 3.4?

------- Additional Comment #6 From Ulrich Drepper 2004-08-08 23:00 -------

I'm building with 3.4 every day.  If somebody has a problem it is with the local
installation.