In ./share/vm/oops/methodData.hpp: we have:

OrderAccess::release_store_ptr(&_cells[index], value);

with no sign of any corresponding load-acquire. This usage is highly suspect. The methodData code has a lot of potential paths to release_store_ptr, and in some places simple release(), combined with some locking. There are no load-acquire uses, in fact there are a couple of comments:

    // No need for "OrderAccess::load_acquire" ops,
    // since the data structure is monotonic.

I do not know what that is intended to mean.

As far as I can see, despite a variety of release_set_xxx methods in the API the only active use is of release_set_bci in RetData::fixup _ret which has the comment:

      // Barrier to ensure displacement is written before the bci; allows
      // the interpreter to read displacement without fear of race condition.

which means that somewhere in the interpreter code there should be a read of a BCI preceding a read of displacement. That read of BCI should be a load-acquire to pair with the use of release_set_bci.