Current PIN/UNPIN directly get translated into a runtime call into GC to pin/unpin the object, which simply push/remove the object reference to/from a root set (synchronisation required). This is inefficient, and also the behaviour is different from the spec (in which you need unpin operations to match every pin operation).

This will need rework, probably along with #12 (closed) . And it is also related with mu-spec#7, which proposes to separate immovability and immortal semantics of PIN instruction.