it isn't clear what instructions are needed by SFS and SFFS in 64-bit mode, I think we should submit a rfc or something to clarify that. e.g. are 128-bit atomics required by 64-bit mode? (it would be nice if they are since gcc/clang enable them by default on powerpc64le-linux-gnu; also since once cache line pinning is implemented, reading/writing 128-bits isn't too horribly difficult) what about bpermd or pdep/pext?
I also think BE-mode should be optional for 64-bit cpus.
imo the best option is to introduce new SFS64 and SFFS64 compliancy levels.