g_assert(cpu->halted) in mttcg_cpu_thread_fn()

[Nicholas Piggin]

I hit this assert in a ppc spapr guest with several CPUs.

There are some paths where cpu1->halted is set to 0 by cpu2, e.g., 
h_prod(). AFAIKS this could trigger the assert if another thread raced 
with the thread that is halting.

Is this just a false positive in the assert, or is our handling of
halted wrong?

Thanks,
Nick