PMU virtualization.

I have to profile apps and OS/drivers in virtualized environment more and more often lately. Usually Vtune timer based sampling / statistical call graph is enough to get the performance data I need.

But some times I have a temptation to continue as I usually do and explore IPC, cache misses, TPB misses, alignment issues, store forwarding, false sharing. Of course it does not work. Most VMs do not support Virtualized PMU. So I assume it is the same on the virtual CPU as when measured on a physical one.

I think now this assumption makes sense, if I understand correctly e.g. what is a meaning of a cache miss in a virtual x86 machine :)