Re: [PATCH] KVM: x86/xen: Fix bug in kvm_xen_vcpu_set_attr()

[Sean Christopherson]

Be more specific in the shortlog.  "Fix a bug in XYZ" doesn't provide any info
about the bug itself, and can even become frustratingly stale if XYZ is renamed.
I believe we should end up with two patches (see below), e.g.

  KVM: x86/xen: Initialize Xen timer only once (when it's NOT running)

and
  
  KVM: x86/xen: Stop Xen timer before changing the IRQ vector

Note, I'm assuming timer_virq is a vector of some form, I haven't actually looked
that far into the code.

On Thu, Jul 28, 2022, Coleman Dietsch wrote:
> This crash appears to be happening when vcpu->arch.xen.timer is already set

Instead of saying "This crash", provide the actual splat (sanitized to make it
more readable).  That way readers, reviewers, and archaeologists don't need to
open up a hyperlink to get details on what broken.

> and kvm_xen_init_timer(vcpu) is called.

Wrap changelogs at ~75 chars.

> During testing with the syzbot reproducer code it seemed apparent that the
> else if statement in the KVM_XEN_VCPU_ATTR_TYPE_TIMER switch case was not
> being reached, which is where the kvm_xen_stop_timer(vcpu) call is located.

Neither the shortlog nor the changelog actually says anything about what is actually
being changed.

> Link: https://syzkaller.appspot.com/bug?id=8234a9dfd3aafbf092cc5a7cd9842e3ebc45fc42
> Reported-and-tested-by: syzbot+e54f930ed78eb0f85281@xxxxxxxxxxxxxxxxxxxxxxxxx
> Signed-off-by: Coleman Dietsch <dietschc@xxxxxxx>
> ---
>  arch/x86/kvm/xen.c | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
> index 610beba35907..4b4b985813c5 100644
> --- a/arch/x86/kvm/xen.c
> +++ b/arch/x86/kvm/xen.c
> @@ -707,6 +707,12 @@ int kvm_xen_vcpu_set_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
>  		break;
>  
>  	case KVM_XEN_VCPU_ATTR_TYPE_TIMER:
> +		/* Stop current timer if it is enabled */
> +		if (kvm_xen_timer_enabled(vcpu)) {
> +			kvm_xen_stop_timer(vcpu);
> +			vcpu->arch.xen.timer_virq = 0;
> +		}
> +
>  		if (data->u.timer.port) {
>  			if (data->u.timer.priority != KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL) {
>  				r = -EINVAL;

I'm not entirely sure this is correct.  Probably doesn't matter, but there's a
subtle ABI change here in that invoking the ioctl with a "bad" priority will
cancel any existing timer.

And there appear to be two separate bugs: initializing the hrtimer while it's
running, and not canceling a running timer before changing timer_virq.

Calling kvm_xen_init_timer() on "every" KVM_XEN_VCPU_ATTR_TYPE_TIMER is odd and
unnecessary, it only needs to be called once during vCPU setup.  If Xen doesn't
have such a hook, then a !ULL check can be done on vcpu->arch.xen.timer.function
to initialize the timer on-demand.

With that out of the way, the code can be streamlined a bit, e.g. something like
this?

	case KVM_XEN_VCPU_ATTR_TYPE_TIMER:
		if (data->u.timer.port &&
		    data->u.timer.priority != KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL) {
			r = -EINVAL;
			break;
		}

		if (!vcpu->arch.xen.timer.function)
			kvm_xen_init_timer(vcpu);

		/* Stop the timer (if it's running) before changing the vector. */
		kvm_xen_stop_timer(vcpu);
		vcpu->arch.xen.timer_virq = data->u.timer.port;

		if (data->u.timer.port && data->u.timer.expires_ns)
			kvm_xen_start_timer(vcpu, data->u.timer.expires_ns,
					    data->u.timer.expires_ns -
					    get_kvmclock_ns(vcpu->kvm));
		r = 0;
		break;

> @@ -720,9 +726,6 @@ int kvm_xen_vcpu_set_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
>  				kvm_xen_start_timer(vcpu, data->u.timer.expires_ns,
>  						    data->u.timer.expires_ns -
>  						    get_kvmclock_ns(vcpu->kvm));
> -		} else if (kvm_xen_timer_enabled(vcpu)) {
> -			kvm_xen_stop_timer(vcpu);
> -			vcpu->arch.xen.timer_virq = 0;
>  		}
>  
>  		r = 0;
> -- 
> 2.34.1
>