The MESI or MOESI protocols need the LLC directory in order to work... and the directory needs to synchronize parallel RFO + snoop-invalidation calls in order for it to work
(in TSO architectures that is... but not necessarily?? IMO a TSO architecture COULD be achieved with an LL/SC based exclusivity acquisition protocol... the ordering flexibility is orthogonal to the cache exclusivity acquisition mechanic... maybe this is just a proprietary choice... maybe not...).
Action which is mostly referred to as "serialize" in the literature... not synchronize (why?).
So, if high-level synchronization (MCS/CLH queues) need MESI... then
What synchronizes the synchronizer?
LLC <-> MMpipeline is the one not getting stalled... while the interaction I'm asking about is theL1x <-> LLC <-> L1ypipeline... in which case I'd argue there must be somebubblinggoing on between L1's and LLC in order to serialize... maybe even some hardware level MCS-like queuing for stores...