Fix support for gigantic stack frames on x86-64

[Geert Bosch]

On Friday, Jun 6, 2003, at 04:11 America/New_York, Richard Henderson 
wrote:
> Well, if you deallocate the stack frame and return in the next
> instruction, there is a 1 insn window where a signal could be
> delivered.  With -fasynchronous-unwind-tables, there are supposed
> to be no such holes.

Of course you could fix this in the unwinder. If the current instruction
pointer points to a return, adjust the instruction pointer and stack 
pointer
as if this return already happened and continue unwinding. I believe 
that
in practice you want the unwinder to be as forgiving as possible for
incomplete/missing information. By always interpreting the effect of a
few known instructions you may be able to correctly unwind most simple
code without descriptors,