KVM: perform a wake_up in kvm_make_all_cpus_request
authorRadim Krčmář <rkrcmar@redhat.com>
Wed, 26 Apr 2017 20:32:23 +0000 (22:32 +0200)
committerPaolo Bonzini <pbonzini@redhat.com>
Thu, 27 Apr 2017 12:14:33 +0000 (14:14 +0200)
We want to have kvm_make_all_cpus_request() to be an optmized version of

  kvm_for_each_vcpu(i, vcpu, kvm) {
    kvm_make_request(vcpu, request);
    kvm_vcpu_kick(vcpu);
  }

and kvm_vcpu_kick() wakes up the target vcpu.  We know which requests do
not need the wake up and use it to optimize the loop.

Thanks to that, this patch doesn't change the behavior of current users
(the all don't need the wake up) and only prepares for future where the
wake up is going to be needed.

I think that most requests do not need the wake up, so we would flip the
bit then.

Later on, kvm_make_request() will take care of kicking too, using this
bit to make the decision whether to kick or not.

Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
virt/kvm/kvm_main.c

index e5d52b46b531d438dcc26fdb49214091ca5bae4c..3772f7dcc72d3f725b19c4abef5cbe980506f89f 100644 (file)
@@ -186,6 +186,9 @@ bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
                /* Set ->requests bit before we read ->mode. */
                smp_mb__after_atomic();
 
+               if (!(req & KVM_REQUEST_NO_WAKEUP))
+                       kvm_vcpu_wake_up(vcpu);
+
                if (cpus != NULL && cpu != -1 && cpu != me &&
                      kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
                        cpumask_set_cpu(cpu, cpus);