2016-03-23 14:18+0100, Paolo Bonzini: > On 23/03/2016 06:08, Yuki Shibuya wrote: >> + /* If nmi pending > 0 and injectable interrupts exist, >> + * nmi pending counter is cleared to prevent skipping >> + * injectable pending interrupts. >> + */ >> + if (vcpu->arch.nmi_pending && kvm_cpu_has_injectable_intr(vcpu) >> + && kvm_x86_ops->interrupt_allowed(vcpu)) >> + vcpu->arch.nmi_pending = 0; > > I am not sure I understand this. Why is it safe to drop nmi_pending? NMIs are latched (queue length 1) and therefore cannot be pending after an injection. I think we want to do it unconditionally. > Can we instead do something like this in vcpu_enter_guest? (We should, even if it doesn't fix the bug. Maskable interrupts can be injected while NMIs are blocked.) What the hell is 2.6.32 doing, though? -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
