KVM: x86: Fix nmi injection failure when vcpu got blocked
authorZhuangYanying <ann.zhuangyanying@huawei.com>
Fri, 26 May 2017 05:16:48 +0000 (13:16 +0800)
committerPaolo Bonzini <pbonzini@redhat.com>
Thu, 1 Jun 2017 09:23:10 +0000 (11:23 +0200)
commit47a66eed99e6f231f4a1d261a9d493f4eee94829
treea7361acd05f5f5a178ab929431add34f29e67307
parentd9c1b5431d5f0e07575db785a022bce91051ac1d
KVM: x86: Fix nmi injection failure when vcpu got blocked

When spin_lock_irqsave() deadlock occurs inside the guest, vcpu threads,
other than the lock-holding one, would enter into S state because of
pvspinlock. Then inject NMI via libvirt API "inject-nmi", the NMI could
not be injected into vm.

The reason is:
1 It sets nmi_queued to 1 when calling ioctl KVM_NMI in qemu, and sets
cpu->kvm_vcpu_dirty to true in do_inject_external_nmi() meanwhile.
2 It sets nmi_queued to 0 in process_nmi(), before entering guest, because
cpu->kvm_vcpu_dirty is true.

It's not enough just to check nmi_queued to decide whether to stay in
vcpu_block() or not. NMI should be injected immediately at any situation.
Add checking nmi_pending, and testing KVM_REQ_NMI replaces nmi_queued
in vm_vcpu_has_events().

Do the same change for SMIs.

Signed-off-by: Zhuang Yanying <ann.zhuangyanying@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/x86.c