preempt_disable() will no longer help prevent CPUs from going offline, once
stop_machine() gets removed from the CPU offline path. So use
get/put_online_cpus_atomic_full() in vmx_vcpu_load() to prevent CPUs from
going offline while clearing vmcs. Here we truly need full-synchronization
with CPU hotplug (and not just an unchanging cpu_online_mask), because we
want to prevent race with the CPU_DYING callback from kvm.

Reported-by: Michael Wang <redacted>
Debugged-by: Xiao Guangrong [off-list ref]
Signed-off-by: Srivatsa S. Bhat <redacted>
---

 arch/x86/kvm/vmx.c |    8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index f858159..23c1063 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c

@@ -1519,10 +1519,14 @@ static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 	struct vcpu_vmx *vmx = to_vmx(vcpu);
 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
 
-	if (!vmm_exclusive)
+	if (!vmm_exclusive) {
 		kvm_cpu_vmxon(phys_addr);
-	else if (vmx->loaded_vmcs->cpu != cpu)
+	} else if (vmx->loaded_vmcs->cpu != cpu) {
+		/* Prevent any CPU from going offline */
+		get_online_cpus_atomic_full();
 		loaded_vmcs_clear(vmx->loaded_vmcs);
+		put_online_cpus_atomic_full();
+	}
 
 	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;