Partitions might introduce timing differences

[francislaus]

Having partitions of microarchitectural state might introduce a timing difference. Depending on which code has run before and thus how many and which partition changes have happened, the timing of the currently running instructions can be impacted. It will run slower or faster depending on whether its partition exists anymore or not. Would this be in contrast to your definition of fence.time?

I have copied in the definition just for reference:
The timing of any instruction or sequence of instructions executing after the fence must be independent of any microarchitectural state before the fence. The flags may exclude this requirement for some subsets of microarchitectural state.

[niwis]

Hi @francislaus, partitions can be allocated and switched statically, or in some other way that does not leak sensitive information. This is up to SW and the system's security policy. Does this answer your question? (Of course, we can also discuss this further at the next TG meeting)

[francislaus]

@niwis I probably have assumptions on my side that differ from what is actually the case. It would be good to learn more about that and use that for a more elaborate answer. Can you point me to more material on what a partition exactly constitutes and how you would want to use it?