This patch does the same thing as the previous one, except it is
complete as-is because it doesn't require architectures to define
TIF_RUNNING. It doesn't use the atomic bitops anymore, and it is
better optimised on UP kernels too.

Affected architectures: ia64, mips, s390, sparc, sparc64, arm.

I have had the patch lightly tested on 2.6.6 on sparc64 SMP,
and different combinations of locking schemes, preempt, smp, etc
tested on i386 SMP.

Comments? Please CC me.

Nick