On Sun, 18 Nov 2018, Linus Torvalds wrote:
This was marked for stable, and honestly, nowhere in the discussion did I see any mention of just *how* bad the performance impact of this was.
When performance goes down by 50% on some loads, people need to start asking themselves whether it was worth it. It's apparently better to just disable SMT entirely, which is what security-conscious people do anyway.
So why do that STIBP slow-down by default when the people who *really* care already disabled SMT?
I think we should use the same logic as for L1TF: we default to something that doesn't kill performance. Warn once about it, and let the crazy people say "I'd rather take a 50% performance hit than worry about a theoretical issue".
Just to update status quo here -- Thomas is working on polishing Tim's set into mergeable state, I've just sent him small addition on top that makes IBPB also be controlled via the same toggle.
That should make the whole 'spectre v2 userspace-to-userspace' mitigation control consistent and undestandable. And also give us even few more % back that are lost due to IBPB as well.