On some guest VMs / containers, sysfs only shows the subset of CPUs allocated for it. Current sysfs code and affinity handling code relies on CPU IDs to be consecutive. On those machines, sysfs parser fails and we fall back to naive topology.

We can make it significantly better.