On Thu, Jun 08, 2017 at 02:50:57PM +0800, Peter Xu wrote:
> On Tue, Jun 06, 2017 at 12:57:05PM +0200, Paolo Bonzini wrote:
> > In some cases, for example involving hot-unplug of assigned
> > devices, pi_post_block can forget to remove the vCPU from the
> > blocked_vcpu_list. When this happens, the next call to
> > pi_pre_block corrupts the list.
> >
> > Fix this in two ways. First, check vcpu->pre_pcpu in pi_pre_block
> > and WARN instead of adding the element twice in the list. Second,
> > always do the list removal in pi_post_block if vcpu->pre_pcpu is
> > set (not -1).
> >
> > The new code keeps interrupts disabled for the whole duration of
> > pi_pre_block/pi_post_block. This is not strictly necessary, but
> > easier to follow. For the same reason, PI.ON is checked only
> > after the cmpxchg, and to handle it we just call the post-block
> > code. This removes duplication of the list removal code.
> >
> > Cc: Longpeng (Mike) <longpe...@huawei.com>
> > Cc: Huangweidong <weidong.hu...@huawei.com>
> > Cc: Gonglei <arei.gong...@huawei.com>
> > Cc: wangxin <wangxinxin.w...@huawei.com>
> > Cc: Radim Krčmář <rkrc...@redhat.com>
> > Signed-off-by: Paolo Bonzini <pbonz...@redhat.com>
> > ---
> > arch/x86/kvm/vmx.c | 62
> > ++++++++++++++++++++++--------------------------------
> > 1 file changed, 25 insertions(+), 37 deletions(-)
> >
> > diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> > index 747d16525b45..0f4714fe4908 100644
> > --- a/arch/x86/kvm/vmx.c
> > +++ b/arch/x86/kvm/vmx.c
> > @@ -11236,10 +11236,11 @@ static void __pi_post_block(struct kvm_vcpu *vcpu)
> > struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
> > struct pi_desc old, new;
> > unsigned int dest;
> > - unsigned long flags;
> >
> > do {
> > old.control = new.control = pi_desc->control;
> > + WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
> > + "Wakeup handler not enabled while the VCPU is blocked\n");
> >
> > dest = cpu_physical_id(vcpu->cpu);
> >
> > @@ -11256,14 +11257,10 @@ static void __pi_post_block(struct kvm_vcpu *vcpu)
> > } while (cmpxchg(&pi_desc->control, old.control,
> > new.control) != old.control);
> >
> > - if(vcpu->pre_pcpu != -1) {
> > - spin_lock_irqsave(
> > - &per_cpu(blocked_vcpu_on_cpu_lock,
> > - vcpu->pre_pcpu), flags);
> > + if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
> > + spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
> > list_del(&vcpu->blocked_vcpu_list);
> > - spin_unlock_irqrestore(
> > - &per_cpu(blocked_vcpu_on_cpu_lock,
> > - vcpu->pre_pcpu), flags);
> > + spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
> > vcpu->pre_pcpu = -1;
> > }
> > }
> > @@ -11283,7 +11280,6 @@ static void __pi_post_block(struct kvm_vcpu *vcpu)
> > */
> > static int pi_pre_block(struct kvm_vcpu *vcpu)
> > {
> > - unsigned long flags;
> > unsigned int dest;
> > struct pi_desc old, new;
> > struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
> > @@ -11293,34 +11289,20 @@ static int pi_pre_block(struct kvm_vcpu *vcpu)
> > !kvm_vcpu_apicv_active(vcpu))
> > return 0;
> >
> > - vcpu->pre_pcpu = vcpu->cpu;
> > - spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
> > - vcpu->pre_pcpu), flags);
> > - list_add_tail(&vcpu->blocked_vcpu_list,
> > - &per_cpu(blocked_vcpu_on_cpu,
> > - vcpu->pre_pcpu));
> > - spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
> > - vcpu->pre_pcpu), flags);
> > + WARN_ON(irqs_disabled());
> > + local_irq_disable();
> > + if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
> > + vcpu->pre_pcpu = vcpu->cpu;
> > + spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
> > + list_add_tail(&vcpu->blocked_vcpu_list,
> > + &per_cpu(blocked_vcpu_on_cpu,
> > + vcpu->pre_pcpu));
> > + spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
> > + }
> >
> > do {
> > old.control = new.control = pi_desc->control;
> >
> > - /*
> > - * We should not block the vCPU if
> > - * an interrupt is posted for it.
> > - */
> > - if (pi_test_on(pi_desc) == 1) {
> > - spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
> > - vcpu->pre_pcpu), flags);
> > - list_del(&vcpu->blocked_vcpu_list);
> > - spin_unlock_irqrestore(
> > - &per_cpu(blocked_vcpu_on_cpu_lock,
> > - vcpu->pre_pcpu), flags);
> > - vcpu->pre_pcpu = -1;
> > -
> > - return 1;
>
> [1]
>
> > - }
> > -
> > WARN((pi_desc->sn == 1),
> > "Warning: SN field of posted-interrupts "
> > "is set before blocking\n");
> > @@ -11345,7 +11327,12 @@ static int pi_pre_block(struct kvm_vcpu *vcpu)
> > } while (cmpxchg(&pi_desc->control, old.control,
> > new.control) != old.control);
> >
> > - return 0;
> > + /* We should not block the vCPU if an interrupt is posted for it. */
> > + if (pi_test_on(pi_desc) == 1)
> > + __pi_post_block(vcpu);
>
> A question on when pi_test_on() is set:
>
> The old code will return 1 if detected (ses [1]), while the new code
> does not. Would that matter? (IIUC that decides whether the vcpu will
> continue to run?)
>
> > +
> > + local_irq_enable();
> > + return (vcpu->pre_pcpu == -1);
>
> Above we have:
>
> if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
> vcpu->pre_pcpu = vcpu->cpu;
> ...
> }
>
> Then can here vcpu->pre_pcpu really be -1?
>
> > }
> >
> > static int vmx_pre_block(struct kvm_vcpu *vcpu)
> > @@ -11361,12 +11348,13 @@ static int vmx_pre_block(struct kvm_vcpu *vcpu)
> >
> > static void pi_post_block(struct kvm_vcpu *vcpu)
> > {
> > - if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
> > - !irq_remapping_cap(IRQ_POSTING_CAP) ||
> > - !kvm_vcpu_apicv_active(vcpu))
> > + if (vcpu->pre_pcpu == -1)
> > return;
> >
> > + WARN_ON(irqs_disabled());
> > + local_irq_disable();
> > __pi_post_block(vcpu);
> > + local_irq_enable();
> > }
> >
> > static void vmx_post_block(struct kvm_vcpu *vcpu)
> > --
> > 2.13.0
> >
> >
>
> A general question to pre_block/post_block handling for PI:
>
> I see that we are handling PI logic mostly in four places:
>
> vmx_vcpu_pi_{load|put}
> pi_{pre_post}_block
>
> But do we really need the pre_block/post_block handling? Here's how I
> understand when vcpu blocked:
>
> - vcpu_block
> - ->pre_block
> - kvm_vcpu_block [1]
> - schedule()
> - kvm_sched_out
> - vmx_vcpu_pi_put [3]
> - (another process working) ...
> - kvm_sched_in
> - vmx_vcpu_pi_load [4]
> - ->post_block [2]
>
> If so, [1] & [2] will definitely be paired with [3] & [4], then why we
> need [3] & [4] at all?
^^^^^^^^^ Here I meant [1] & [2]. Sorry.
>
> (Though [3] & [4] will also be used when preemption happens, so they
> are required)
>
> Please kindly figure out if I missed anything important...
>
> Thanks,
>
> --
> Peter Xu
--
Peter Xu
