On Thu, Jun 01, 2017 at 02:37:52PM +0200, Paolo Bonzini wrote: > > > On 01/06/2017 13:09, Andrew Jones wrote: > >>> - kvm_vcpu_kick(vcpu); > >>> + swake_up(kvm_arch_vcpu_wq(vcpu)); > >> We have kvm_vcpu_wake_up(). Why not use that? > > The are two differences between swake_up(kvm_arch_vcpu_wq(vcpu)) and > > kvm_vcpu_wake_up(vcpu) > > 1. kvm_vcpu_wake_up() has a return value: true on wake up, else false > > 2. kvm_vcpu_wake_up() increments the halt_wakeup stat when the vcpu > > is awaken > > > > (1) doesn't really matter, but (2) might. Hmm, I think we do want to > > increment that stat in this case though, so I should change this. > > Yep. > > > Also, we have another use of swake_up(kvm_arch_vcpu_wq(vcpu)), in > > kvm_arm_resume_guest(), but there I don't think we want to increment > > the halt stat, so that one is probably OK. > > I would define a __kvm_vcpu_wake_up if you don't want the stat. > I didn't mind the other one, because it is in a function in the code right above the sleep function, which uses swait_event_interruptible directly without calling kvm_vcpu_block, so the symmetry is clear there, but not in the timer case. -Christoffer
