ForeignPtr and cfinalizer are essential for handling c++ objects with ghc. However using cfinalizer cuts the throughput in half. ghc version is 9.4.2. https://github.com/junjihashimoto/ffi-benchmark In this example, ForeignPtr_ without cfinalizer takes 60ns, and ForeignPtr with cfinalizer takes 150ns. I tried to eliminate the influence of the internal weak pointer lock, but it(CForeignPtr without lock) takes 130ns. I think there is also an influence of gc, but is it possible to reduce this overhead? This is an image capture of the criterion in the above repository. ForeignPtr_ does not use cfinalizer. It's a typo.