Re: [PATCH] KVM: arm64: Mark the VM as dead for failed initializations

[Oliver Upton <oliver.upton@linux.dev>]

On Sat, Oct 26, 2024 at 08:43:21AM +0100, Marc Zyngier wrote:
> diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> index bf64fed9820e..c315bc1a4e9a 100644
> --- a/arch/arm64/include/asm/kvm_host.h
> +++ b/arch/arm64/include/asm/kvm_host.h
> @@ -74,8 +74,6 @@ enum kvm_mode kvm_get_mode(void);
>  static inline enum kvm_mode kvm_get_mode(void) { return KVM_MODE_NONE; };
>  #endif
>  
> -DECLARE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
> -
>  extern unsigned int __ro_after_init kvm_sve_max_vl;
>  extern unsigned int __ro_after_init kvm_host_sve_max_vl;
>  int __init kvm_arm_init_sve(void);
> diff --git a/arch/arm64/kvm/arch_timer.c b/arch/arm64/kvm/arch_timer.c
> index 879982b1cc73..1215df590418 100644
> --- a/arch/arm64/kvm/arch_timer.c
> +++ b/arch/arm64/kvm/arch_timer.c
> @@ -206,8 +206,7 @@ void get_timer_map(struct kvm_vcpu *vcpu, struct timer_map *map)
>  
>  static inline bool userspace_irqchip(struct kvm *kvm)
>  {
> -	return static_branch_unlikely(&userspace_irqchip_in_use) &&
> -		unlikely(!irqchip_in_kernel(kvm));
> +	return unlikely(!irqchip_in_kernel(kvm));
>  }
>  
>  static void soft_timer_start(struct hrtimer *hrt, u64 ns)
> diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
> index 48cafb65d6ac..70ff9a20ef3a 100644
> --- a/arch/arm64/kvm/arm.c
> +++ b/arch/arm64/kvm/arm.c
> @@ -69,7 +69,6 @@ DECLARE_KVM_NVHE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
>  static bool vgic_present, kvm_arm_initialised;
>  
>  static DEFINE_PER_CPU(unsigned char, kvm_hyp_initialized);
> -DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
>  
>  bool is_kvm_arm_initialised(void)
>  {
> @@ -503,9 +502,6 @@ void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
>  
>  void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
>  {
> -	if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
> -		static_branch_dec(&userspace_irqchip_in_use);
> -
>  	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
>  	kvm_timer_vcpu_terminate(vcpu);
>  	kvm_pmu_vcpu_destroy(vcpu);
> @@ -848,14 +844,6 @@ int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
>  			return ret;
>  	}
>  
> -	if (!irqchip_in_kernel(kvm)) {
> -		/*
> -		 * Tell the rest of the code that there are userspace irqchip
> -		 * VMs in the wild.
> -		 */
> -		static_branch_inc(&userspace_irqchip_in_use);
> -	}
> -
>  	/*
>  	 * Initialize traps for protected VMs.
>  	 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
> @@ -1077,7 +1065,7 @@ static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
>  	 * state gets updated in kvm_timer_update_run and
>  	 * kvm_pmu_update_run below).
>  	 */
> -	if (static_branch_unlikely(&userspace_irqchip_in_use)) {
> +	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
>  		if (kvm_timer_should_notify_user(vcpu) ||
>  		    kvm_pmu_should_notify_user(vcpu)) {
>  			*ret = -EINTR;
> @@ -1199,7 +1187,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
>  			vcpu->mode = OUTSIDE_GUEST_MODE;
>  			isb(); /* Ensure work in x_flush_hwstate is committed */
>  			kvm_pmu_sync_hwstate(vcpu);
> -			if (static_branch_unlikely(&userspace_irqchip_in_use))
> +			if (unlikely(!irqchip_in_kernel(vcpu->kvm)))
>  				kvm_timer_sync_user(vcpu);
>  			kvm_vgic_sync_hwstate(vcpu);
>  			local_irq_enable();
> @@ -1245,7 +1233,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
>  		 * we don't want vtimer interrupts to race with syncing the
>  		 * timer virtual interrupt state.
>  		 */
> -		if (static_branch_unlikely(&userspace_irqchip_in_use))
> +		if (unlikely(!irqchip_in_kernel(vcpu->kvm)))
>  			kvm_timer_sync_user(vcpu);
>  
>  		kvm_arch_vcpu_ctxsync_fp(vcpu);
> 
> I think this would fix the problem you're seeing without changing the
> userspace view of an erroneous configuration. It would also pave the
> way for the complete removal of the interrupt notification to
> userspace, which I claim has no user and is just a shit idea.

Yeah, looks like this ought to get it done.

Even with a fix for this particular issue I do wonder if we should
categorically harden against late initialization failures and un-init
the vCPU (or bug VM, where necessary) to avoid dealing with half-baked
vCPUs/VMs across our UAPI surfaces.

A sane userspace will probably crash when KVM_RUN returns EINVAL anyway.

-- 
Thanks,
Oliver