KVM: avoid unnecessary wait for a async pf
authorXiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Mon, 1 Nov 2010 09:01:28 +0000 (17:01 +0800)
committerAvi Kivity <avi@redhat.com>
Wed, 12 Jan 2011 09:29:00 +0000 (11:29 +0200)
In current code, it checks async pf completion out of the wait context,
like this:

if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
    !vcpu->arch.apf.halted)
r = vcpu_enter_guest(vcpu);
else {
......
kvm_vcpu_block(vcpu)
 ^- waiting until 'async_pf.done' is not empty
}

kvm_check_async_pf_completion(vcpu)
 ^- delete list from async_pf.done

So, if we check aysnc pf completion first, it can be blocked at
kvm_vcpu_block

Fixed by mark the vcpu is unhalted in kvm_check_async_pf_completion()
path

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Acked-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
arch/x86/kvm/x86.c

index ab10a6ca154927c7b724544207b4122f5737462e..5daead8336690206f64431b7169c6abe043280cb 100644 (file)
@@ -6283,6 +6283,7 @@ void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
                vcpu->arch.fault.address = work->arch.token;
                kvm_inject_page_fault(vcpu);
        }
+       vcpu->arch.apf.halted = false;
 }
 
 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)