For hazard pointers it was proposed here https://groups.google.com/g/comp.programming.threads/c/XU6Bt...

I did a POC afterwards. For linux it was a bit PITA since /proc text had to be parsed to track context switches w/ associated full MB. For Mac OS on the ppc also bit PITA since Apple tried to hide their unix api and some stuff involved calling the mach micro kernel.

An informal proof of why memory barrierless hazard pointers can't have false negatives. https://drive.google.com/file/d/1zDrXDdJHQEYILlwUSILPfSSiVzM...