Re: [PATCH v3 07/26] KVM: VMX: Move preemption timer <=> hrtimer dance to common x86

[Paolo Bonzini]

On 3/29/23 14:34, Tudor Ambarus wrote:
> This patch fixes the bug reported at:
> LINK:
> https://syzkaller.appspot.com/bug?id=489beb3d76ef14cc6cd18125782dc6f86051a605
>
> One may find the strace at:
> LINK:https://syzkaller.appspot.com/text?tag=CrashLog&x=1798b54ec80000
> and the c reproducer at:
> LINK:https://syzkaller.appspot.com/text?tag=ReproC&x=10365781c80000
>
> Since I've no experience with kvm, it would be helpful if one of you can
> provide some guidance. Do you think it is worth to backport this patch
> to stable (together with its prerequisite patches), or shall I try to
> get familiar with the code and try to provide a less invasive fix?

I think it is enough to fix the conflicts in vmx_pre_block and
vmx_post_block, there are no prerequisites:

diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index 0718658268fe..895069038856 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -7577,17 +7577,11 @@ static int vmx_pre_block(struct kvm_vcpu *vcpu)
 	if (pi_pre_block(vcpu))
 		return 1;
 
-	if (kvm_lapic_hv_timer_in_use(vcpu))
-		kvm_lapic_switch_to_sw_timer(vcpu);
-
 	return 0;
 }
 
 static void vmx_post_block(struct kvm_vcpu *vcpu)
 {
-	if (kvm_x86_ops.set_hv_timer)
-		kvm_lapic_switch_to_hv_timer(vcpu);
-
 	pi_post_block(vcpu);
 }
 
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index fcfa3fedf84f..4eca3ec38afd 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -10022,12 +10022,28 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
 
 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
 {
+	bool hv_timer;
+
 	if (!kvm_arch_vcpu_runnable(vcpu) &&
 	    (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
+		/*
+		 * Switch to the software timer before halt-polling/blocking as
+		 * the guest's timer may be a break event for the vCPU, and the
+		 * hypervisor timer runs only when the CPU is in guest mode.
+		 * Switch before halt-polling so that KVM recognizes an expired
+		 * timer before blocking.
+		 */
+		hv_timer = kvm_lapic_hv_timer_in_use(vcpu);
+		if (hv_timer)
+			kvm_lapic_switch_to_sw_timer(vcpu);
+
 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
 		kvm_vcpu_block(vcpu);
 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
 
+		if (hv_timer)
+			kvm_lapic_switch_to_hv_timer(vcpu);
+
 		if (kvm_x86_ops.post_block)
 			static_call(kvm_x86_post_block)(vcpu);
 
@@ -10266,6 +10282,11 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
 			r = -EINTR;
 			goto out;
 		}
+		/*
+		 * It should be impossible for the hypervisor timer to be in
+		 * use before KVM has ever run the vCPU.
+		 */
+		WARN_ON_ONCE(kvm_lapic_hv_timer_in_use(vcpu));
 		kvm_vcpu_block(vcpu);
 		if (kvm_apic_accept_events(vcpu) < 0) {
 			r = 0;

The fix is due to the second "if" changing from
kvm_x86_ops.set_hv_timer to hv_timer.

Paolo