KVM: x86: Avoid issuing wbinvd twice
authorJan Kiszka <jan.kiszka@siemens.com>
Mon, 1 Nov 2010 13:01:29 +0000 (14:01 +0100)
committerAvi Kivity <avi@redhat.com>
Wed, 12 Jan 2011 09:28:52 +0000 (11:28 +0200)
Micro optimization to avoid calling wbinvd twice on the CPU that has to
emulate it. As we might be preempted between smp_call_function_many and
the local wbinvd, the cache might be filled again so that real work
could be done uselessly.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
arch/x86/kvm/x86.c

index 35f82f2c66f6397e3a329d46f1adf82b7ac7fa99..c10135bc0f2fd9580380411158806c5f9af05ffa 100644 (file)
@@ -4007,13 +4007,15 @@ int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
                return X86EMUL_CONTINUE;
 
        if (kvm_x86_ops->has_wbinvd_exit()) {
-               preempt_disable();
+               int cpu = get_cpu();
+
+               cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
                smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
                                wbinvd_ipi, NULL, 1);
-               preempt_enable();
+               put_cpu();
                cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
-       }
-       wbinvd();
+       } else
+               wbinvd();
        return X86EMUL_CONTINUE;
 }
 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);