Currently, uw_init_context_1 will always call init_dwarf_reg_size_table
if __GTHREADS is defined but threading is not active or not actually
implemented.  The stub functions for __gthread_once return -1
to indicate that this functionality is unimplemented.  Likewise, full
implementations return -1 when threadig is not active.
We will currently call init_dwarf_reg_size_table unguarded if either
__gthread_once fails or dwarf_reg_size_table[0] is zero; this is bogus.

If _gthread_once succeeds, we know that init_dwarf_reg_size_table
has been called, so there is no need to test dwarf_reg_size_table[0],
we definitely don't need to call it again.
If it fails, this can be because threading is not active, so we should
test dwarf_reg_size_table[0].  If that is set, it indicates that
we've already called init_dwarf_reg_size_table, just like in the
!__GTHREADS case.

I've successfully regression tested the below patch on
i686-pc-linux-gnu X sh-elf, however, that doesn't really mean all that
much since this is currently not a __GTHREADS target.
I can't bootstrap mainline on my host, see
http://gcc.gnu.org/ml/gcc/2006-04/msg00055.html .
Could someone with a __GTHREADS host which can still bootstrap test this
patch?

2006-04-05  J"orn Rennecke <joern.rennecke@st.com>

	* unwind-dw2.c (uw_init_context_1): Fix test that checks if
	dwarf_reg_size_table has been initialized.

Index: unwind-dw2.c
===================================================================
/usr/bin/diff -p -d -F^( -u -L unwind-dw2.c	(revision 112638) -L unwind-dw2.c	(working copy) .svn/text-base/unwind-dw2.c.svn-base unwind-dw2.c
--- unwind-dw2.c	(revision 112638)
+++ unwind-dw2.c	(working copy)
@@ -1370,7 +1370,7 @@ uw_init_context_1 (struct _Unwind_Contex
   {
     static __gthread_once_t once_regsizes = __GTHREAD_ONCE_INIT;
     if (__gthread_once (&once_regsizes, init_dwarf_reg_size_table) != 0
-	|| dwarf_reg_size_table[0] == 0)
+	&& dwarf_reg_size_table[0] == 0)
       init_dwarf_reg_size_table ();
   }
 #else