The calculation of the tsc_delta value to ensure a
forward-going tsc for the guest is a function of the
host-tsc. This works as long as the guests tsc_khz is equal
to the hosts tsc_khz. With tsc-scaling hardware support this
is not longer true and the tsc_delta needs to be calculated
using guest_tsc values.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
        kvm_x86_ops->vcpu_load(vcpu, cpu);
        if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
                /* Make sure TSC doesn't go backwards */
-               s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
-                               native_read_tsc() - vcpu->arch.last_host_tsc;
+               s64 tsc_delta;
+               u64 tsc;
+
+               kvm_get_msr(vcpu, MSR_IA32_TSC, &tsc);
+               tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
+                            tsc - vcpu->arch.last_guest_tsc;
+
                if (tsc_delta < 0)
                        mark_tsc_unstable("KVM discovered backwards TSC");
                if (check_tsc_unstable()) {