Two problems with existing code:
1) Grows the simulated memory one byte at a time, and
2) If target_mem.base ever moves down, doesn't compensate
by moving the contents up.