L2 was running with uninitialized PML fields which led to incomplete
dirty bitmap logging. This manifested as all kinds of subtle erratic
behavior of the nested guest.
Fixes:
843e4330573c ("KVM: VMX: Add PML support in VMX")
Signed-off-by: Ladi Prosek <lprosek@redhat.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
}
+ if (enable_pml) {
+ /*
+ * Conceptually we want to copy the PML address and index from
+ * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
+ * since we always flush the log on each vmexit, this happens
+ * to be equivalent to simply resetting the fields in vmcs02.
+ */
+ ASSERT(vmx->pml_pg);
+ vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
+ vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
+ }
+
if (nested_cpu_has_ept(vmcs12)) {
kvm_mmu_unload(vcpu);
nested_ept_init_mmu_context(vcpu);