I am trying to profile an application running on a qemu-kvm VM. My host machine is a KabyLake processor with VT-x extensions. I am using perf-tool inside the VM to read the performance counters (by passing the -cpu=host flag to qemu and using the Guest PMU)
Regarding the TLB miss performance counters I have the following question:
Do the DTLB_LOAD_MISSES.WALK_ACTIVE and DTLB_STORE_MISSES.WALK_ACTIVE counters exclude the cycles spent walking the Extended Page Tables (EPT)? If this is the case to find out the total amount of cycles that the application spent on page walking (both guest page tables and extended page tables) I should add all counters: DTLB_LOAD_MISSES.WALK_ACTIVE + DTLB_STORE_MISSES.WALK_ACTIVE + EPT.WALK_PENDING?