This serves to highlight the need to return to a model of computing where the call stack is not addressable by ordinary applications.

The 'display' where arguments and sometimes return values are kept can continue to be done as it is now. It's just the return addresses to calls? They are the root of many of the issues plaguing cpu security today.

If a separate, un-addressable (except by call and return instructions, indirectly) stack would do much to mitigate Spectre, ROP-oriented hacks and so on.