I think it's worth noting that requests come from the last level cache as opposed to the processor. Does this mean that the lower level cache will always be a superset of the higher level cache? How does write-back write through coherency policies come into play?