Description of problem:


The KVM-RT team saw large cyclictest latency's during testing of RT kernels
that were built with the recent set of CVEs.

One of these CVEs, CVE-2018-1220, introduced code for recovering
shattered huge pages in the kvm/mmu.c code. A settable
variable in this code, nx_huge_pages_recovery_ratio, was defaulted to
the value 60, which means that the recovery code would loop potentially
60 times while the kvm->mmu_lock was held. This bit of code introduced
very large (> 5000us) on single VMs with multiple vCPUs. On a single VM with 1
vCPU, the max latency was around 138us which is still unacceptable. 

Version-Release number of selected component (if applicable):


How reproducible:


Steps to Reproduce:
1.
2.
3.

Actual results:


Expected results:


Additional info: