[patch 0/2] timer injection races

[patch 0/2] timer injection races

[Marcelo Tosatti]

-- 

[patch 1/2] KVM: consolidate check for pending vcpu requests

[Marcelo Tosatti]

[-- Attachment #1: better-vcpu-run-request-handling --]
[-- Type: text/plain, Size: 1659 bytes --]

A guest vcpu instance can be scheduled to a different physical CPU
between the test for KVM_REQ_MIGRATE_TIMER and local_irq_disable().

If that happens, the timer will only be migrated to the current pCPU on
the next exit, meaning that guest LAPIC timer event can be delayed until
a host interrupt is triggered.

Fix it by cancelling guest entry if any vcpu request is pending.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: kvm/arch/x86/kvm/x86.c
===================================================================
--- kvm.orig/arch/x86/kvm/x86.c
+++ kvm/arch/x86/kvm/x86.c
@@ -2798,6 +2798,8 @@ again:
 	if (vcpu->requests) {
 		if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
 			__kvm_migrate_timers(vcpu);
+		if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
+			kvm_x86_ops->tlb_flush(vcpu);
 		if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
 				       &vcpu->requests)) {
 			kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
@@ -2820,21 +2822,13 @@ again:
 
 	local_irq_disable();
 
-	if (need_resched()) {
+	if (vcpu->requests || need_resched()) {
 		local_irq_enable();
 		preempt_enable();
 		r = 1;
 		goto out;
 	}
 
-	if (vcpu->requests)
-		if (test_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests)) {
-			local_irq_enable();
-			preempt_enable();
-			r = 1;
-			goto out;
-		}
-
 	if (signal_pending(current)) {
 		local_irq_enable();
 		preempt_enable();
@@ -2864,9 +2858,6 @@ again:
 
 	kvm_guest_enter();
 
-	if (vcpu->requests)
-		if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
-			kvm_x86_ops->tlb_flush(vcpu);
 
 	KVMTRACE_0D(VMENTRY, vcpu, entryexit);
 	kvm_x86_ops->run(vcpu, kvm_run);

-- 

Re: [patch 1/2] KVM: consolidate check for pending vcpu requests

[Avi Kivity]

Marcelo Tosatti wrote:
> A guest vcpu instance can be scheduled to a different physical CPU
> between the test for KVM_REQ_MIGRATE_TIMER and local_irq_disable().
>
> If that happens, the timer will only be migrated to the current pCPU on
> the next exit, meaning that guest LAPIC timer event can be delayed until
> a host interrupt is triggered.
>
> Fix it by cancelling guest entry if any vcpu request is pending.
>   

Applied, thanks.

-- 
error compiling committee.c: too many arguments to function

[patch 2/2] KVM: close timer injection race window in __vcpu_run

[Marcelo Tosatti]

[-- Attachment #1: fired-but-not-injected-guest-timer --]
[-- Type: text/plain, Size: 2350 bytes --]

If a timer fires after kvm_inject_pending_timer_irqs() but before
local_irq_disable() the code will enter guest mode and only inject such
timer interrupt the next time an unrelated event causes an exit.

It would be simpler if the timer->pending irq conversion could be done
with IRQ's disabled, so that the above problem cannot happen.

For now introduce a new vcpu requests bit to cancel guest entry.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: kvm/arch/x86/kvm/i8254.c
===================================================================
--- kvm.orig/arch/x86/kvm/i8254.c
+++ kvm/arch/x86/kvm/i8254.c
@@ -200,9 +200,12 @@ static int __pit_timer_fn(struct kvm_kpi
 
 	atomic_inc(&pt->pending);
 	smp_mb__after_atomic_inc();
-	if (vcpu0 && waitqueue_active(&vcpu0->wq)) {
-		vcpu0->arch.mp_state = KVM_MP_STATE_RUNNABLE;
-		wake_up_interruptible(&vcpu0->wq);
+	if (vcpu0) {
+		set_bit(KVM_REQ_PENDING_TIMER, &vcpu0->requests);
+		if (waitqueue_active(&vcpu0->wq)) {
+			vcpu0->arch.mp_state = KVM_MP_STATE_RUNNABLE;
+			wake_up_interruptible(&vcpu0->wq);
+		}
 	}
 
 	pt->timer.expires = ktime_add_ns(pt->timer.expires, pt->period);
Index: kvm/arch/x86/kvm/lapic.c
===================================================================
--- kvm.orig/arch/x86/kvm/lapic.c
+++ kvm/arch/x86/kvm/lapic.c
@@ -946,6 +946,7 @@ static int __apic_timer_fn(struct kvm_la
 	wait_queue_head_t *q = &apic->vcpu->wq;
 
 	atomic_inc(&apic->timer.pending);
+	set_bit(KVM_REQ_PENDING_TIMER, &apic->vcpu->requests);
 	if (waitqueue_active(q)) {
 		apic->vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
 		wake_up_interruptible(q);
Index: kvm/arch/x86/kvm/x86.c
===================================================================
--- kvm.orig/arch/x86/kvm/x86.c
+++ kvm/arch/x86/kvm/x86.c
@@ -2813,6 +2813,7 @@ again:
 		}
 	}
 
+	clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
 	kvm_inject_pending_timer_irqs(vcpu);
 
 	preempt_disable();
Index: kvm/include/linux/kvm_host.h
===================================================================
--- kvm.orig/include/linux/kvm_host.h
+++ kvm/include/linux/kvm_host.h
@@ -33,6 +33,7 @@
 #define KVM_REQ_REPORT_TPR_ACCESS  2
 #define KVM_REQ_MMU_RELOAD         3
 #define KVM_REQ_TRIPLE_FAULT       4
+#define KVM_REQ_PENDING_TIMER      5
 
 struct kvm_vcpu;
 extern struct kmem_cache *kvm_vcpu_cache;

-- 

Re: [patch 2/2] KVM: close timer injection race window in __vcpu_run

[Avi Kivity]

Marcelo Tosatti wrote:
> If a timer fires after kvm_inject_pending_timer_irqs() but before
> local_irq_disable() the code will enter guest mode and only inject such
> timer interrupt the next time an unrelated event causes an exit.
>
> It would be simpler if the timer->pending irq conversion could be done
> with IRQ's disabled, so that the above problem cannot happen.
>
> For now introduce a new vcpu requests bit to cancel guest entry.
>
> Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
>
>   

Applied this too.

> Index: kvm/arch/x86/kvm/i8254.c
> ===================================================================
> --- kvm.orig/arch/x86/kvm/i8254.c
> +++ kvm/arch/x86/kvm/i8254.c
> @@ -200,9 +200,12 @@ static int __pit_timer_fn(struct kvm_kpi
>  
>  	atomic_inc(&pt->pending);
>  	smp_mb__after_atomic_inc();
> -	if (vcpu0 && waitqueue_active(&vcpu0->wq)) {
> -		vcpu0->arch.mp_state = KVM_MP_STATE_RUNNABLE;
> -		wake_up_interruptible(&vcpu0->wq);
> +	if (vcpu0) {
> +		set_bit(KVM_REQ_PENDING_TIMER, &vcpu0->requests);
> +		if (waitqueue_active(&vcpu0->wq)) {
> +			vcpu0->arch.mp_state = KVM_MP_STATE_RUNNABLE;
> +			wake_up_interruptible(&vcpu0->wq);
> +		}
>  	}
>  
>   

We probably ought to wakeup only if pt->pending was zero, no?

-- 
error compiling committee.c: too many arguments to function

Re: [patch 2/2] KVM: close timer injection race window in __vcpu_run

[Marcelo Tosatti]

On Sun, Jun 08, 2008 at 10:17:15AM +0300, Avi Kivity wrote:
>> Index: kvm/arch/x86/kvm/i8254.c
>> ===================================================================
>> --- kvm.orig/arch/x86/kvm/i8254.c
>> +++ kvm/arch/x86/kvm/i8254.c
>> @@ -200,9 +200,12 @@ static int __pit_timer_fn(struct kvm_kpi
>>   	atomic_inc(&pt->pending);
>>  	smp_mb__after_atomic_inc();
>> -	if (vcpu0 && waitqueue_active(&vcpu0->wq)) {
>> -		vcpu0->arch.mp_state = KVM_MP_STATE_RUNNABLE;
>> -		wake_up_interruptible(&vcpu0->wq);
>> +	if (vcpu0) {
>> +		set_bit(KVM_REQ_PENDING_TIMER, &vcpu0->requests);
>> +		if (waitqueue_active(&vcpu0->wq)) {
>> +			vcpu0->arch.mp_state = KVM_MP_STATE_RUNNABLE;
>> +			wake_up_interruptible(&vcpu0->wq);
>> +		}
>>  	}
>>    
>
> We probably ought to wakeup only if pt->pending was zero, no?

Yep, same for LAPIC. 

Re: [patch 2/2] KVM: close timer injection race window in __vcpu_run

[Marcelo Tosatti]

On Sun, Jun 08, 2008 at 12:08:34PM -0300, Marcelo Tosatti wrote:
> On Sun, Jun 08, 2008 at 10:17:15AM +0300, Avi Kivity wrote:
> >
> > We probably ought to wakeup only if pt->pending was zero, no?
> 
> Yep, same for LAPIC. 

Using atomic_inc_and_test() for it also introduces an SMP barrier
to the LAPIC version (thought it was unecessary because of timer
migration, but guest can be scheduled to a different pCPU between exit
and kvm_vcpu_block(), so there is the possibility for a race).

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>


diff --git a/arch/x86/kvm/i8254.c b/arch/x86/kvm/i8254.c
index 9e3391e..c0f7872 100644
--- a/arch/x86/kvm/i8254.c
+++ b/arch/x86/kvm/i8254.c
@@ -198,14 +198,11 @@ static int __pit_timer_fn(struct kvm_kpit_state *ps)
 	struct kvm_vcpu *vcpu0 = ps->pit->kvm->vcpus[0];
 	struct kvm_kpit_timer *pt = &ps->pit_timer;
 
-	atomic_inc(&pt->pending);
-	smp_mb__after_atomic_inc();
-	if (vcpu0) {
+	if (!atomic_inc_and_test(&pt->pending))
 		set_bit(KVM_REQ_PENDING_TIMER, &vcpu0->requests);
-		if (waitqueue_active(&vcpu0->wq)) {
-			vcpu0->arch.mp_state = KVM_MP_STATE_RUNNABLE;
-			wake_up_interruptible(&vcpu0->wq);
-		}
+	if (vcpu0 && waitqueue_active(&vcpu0->wq)) {
+		vcpu0->arch.mp_state = KVM_MP_STATE_RUNNABLE;
+		wake_up_interruptible(&vcpu0->wq);
 	}
 
 	pt->timer.expires = ktime_add_ns(pt->timer.expires, pt->period);
diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index 180ba73..73f43de 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -945,8 +945,8 @@ static int __apic_timer_fn(struct kvm_lapic *apic)
 	int result = 0;
 	wait_queue_head_t *q = &apic->vcpu->wq;
 
-	atomic_inc(&apic->timer.pending);
-	set_bit(KVM_REQ_PENDING_TIMER, &apic->vcpu->requests);
+	if(!atomic_inc_and_test(&apic->timer.pending))
+		set_bit(KVM_REQ_PENDING_TIMER, &apic->vcpu->requests);
 	if (waitqueue_active(q)) {
 		apic->vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
 		wake_up_interruptible(q);