Add a comment to explain why KVM clears IsRunning when putting a vCPU, even though leaving IsRunning=1 would be ok from a functional perspective. Per Maxim's experiments, a misbehaving VM could spam the AVIC doorbell so fast as to induce a 50%+ loss in performance. Link: https://lore.kernel.org/all/8d7e0d0391df4efc7cb28557297eb2ec9904f1e5.camel@xxxxxxxxxx Cc: Maxim Levitsky <mlevitsk@xxxxxxxxxx> Signed-off-by: Sean Christopherson <seanjc@xxxxxxxxxx> --- arch/x86/kvm/svm/avic.c | 31 ++++++++++++++++++------------- 1 file changed, 18 insertions(+), 13 deletions(-) diff --git a/arch/x86/kvm/svm/avic.c b/arch/x86/kvm/svm/avic.c index aba3f9d2ad02..60e6e82fe41f 100644 --- a/arch/x86/kvm/svm/avic.c +++ b/arch/x86/kvm/svm/avic.c @@ -1133,19 +1133,24 @@ void avic_vcpu_blocking(struct kvm_vcpu *vcpu) if (!kvm_vcpu_apicv_active(vcpu)) return; - /* - * Unload the AVIC when the vCPU is about to block, _before_ - * the vCPU actually blocks. - * - * Any IRQs that arrive before IsRunning=0 will not cause an - * incomplete IPI vmexit on the source, therefore vIRR will also - * be checked by kvm_vcpu_check_block() before blocking. The - * memory barrier implicit in set_current_state orders writing - * IsRunning=0 before reading the vIRR. The processor needs a - * matching memory barrier on interrupt delivery between writing - * IRR and reading IsRunning; the lack of this barrier might be - * the cause of errata #1235). - */ + /* + * Unload the AVIC when the vCPU is about to block, _before_ the vCPU + * actually blocks. + * + * Note, any IRQs that arrive before IsRunning=0 will not cause an + * incomplete IPI vmexit on the source; kvm_vcpu_check_block() handles + * this by checking vIRR one last time before blocking. The memory + * barrier implicit in set_current_state orders writing IsRunning=0 + * before reading the vIRR. The processor needs a matching memory + * barrier on interrupt delivery between writing IRR and reading + * IsRunning; the lack of this barrier might be the cause of errata #1235). + * + * Clear IsRunning=0 even if guest IRQs are disabled, i.e. even if KVM + * doesn't need to detect events for scheduling purposes. The doorbell + * used to signal running vCPUs cannot be blocked, i.e. will perturb the + * CPU and cause noisy neighbor problems if the VM is sending interrupts + * to the vCPU while it's scheduled out. + */ avic_vcpu_put(vcpu); } -- 2.49.0.504.g3bcea36a83-goog
