[PATCH v2 4/4] KVM: arm64: Create each pKVM hyp vcpu after its corresponding host vcpu

[Marc Zyngier]

On Wed, 26 Feb 2025 21:55:20 +0000,
Fuad Tabba <tabba at google.com> wrote:
> 
> Instead of creating and initializing _all_ hyp vcpus in pKVM when
> the first host vcpu runs for the first time, initialize _each_
> hyp vcpu in conjunction with its corresponding host vcpu.
> 
> Some of the host vcpu state (e.g., system registers and traps
> values) is not initialized until the first time the host vcpu is
> run. Therefore, initializing a hyp vcpu before its corresponding
> host vcpu has run for the first time might not view the complete
> host state of these vcpus.
> 
> Additionally, this behavior is inline with non-protected modes.
> 
> Signed-off-by: Fuad Tabba <tabba at google.com>
> ---
>  arch/arm64/include/asm/kvm_host.h      |  2 +
>  arch/arm64/include/asm/kvm_pkvm.h      |  1 +
>  arch/arm64/kvm/arm.c                   |  4 ++
>  arch/arm64/kvm/hyp/include/nvhe/pkvm.h |  6 ---
>  arch/arm64/kvm/hyp/nvhe/pkvm.c         | 54 +++++++++++++++-----------
>  arch/arm64/kvm/pkvm.c                  | 28 ++++++-------
>  6 files changed, 53 insertions(+), 42 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> index 3a7ec98ef123..4a0e522f6001 100644
> --- a/arch/arm64/include/asm/kvm_host.h
> +++ b/arch/arm64/include/asm/kvm_host.h
> @@ -869,6 +869,8 @@ struct kvm_vcpu_arch {
>  #define VCPU_INITIALIZED	__vcpu_single_flag(cflags, BIT(0))
>  /* SVE config completed */
>  #define VCPU_SVE_FINALIZED	__vcpu_single_flag(cflags, BIT(1))
> +/* pKVM VCPU setup completed */
> +#define VCPU_PKVM_FINALIZED	__vcpu_single_flag(cflags, BIT(2))
>  
>  /* Exception pending */
>  #define PENDING_EXCEPTION	__vcpu_single_flag(iflags, BIT(0))
> diff --git a/arch/arm64/include/asm/kvm_pkvm.h b/arch/arm64/include/asm/kvm_pkvm.h
> index eb65f12e81d9..abd693ce5b93 100644
> --- a/arch/arm64/include/asm/kvm_pkvm.h
> +++ b/arch/arm64/include/asm/kvm_pkvm.h
> @@ -19,6 +19,7 @@
>  int pkvm_init_host_vm(struct kvm *kvm);
>  int pkvm_create_hyp_vm(struct kvm *kvm);
>  void pkvm_destroy_hyp_vm(struct kvm *kvm);
> +int pkvm_create_hyp_vcpu(struct kvm_vcpu *vcpu);
>  
>  /*
>   * This functions as an allow-list of protected VM capabilities.
> diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
> index b8e55a441282..cd7a808eeb64 100644
> --- a/arch/arm64/kvm/arm.c
> +++ b/arch/arm64/kvm/arm.c
> @@ -833,6 +833,10 @@ int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
>  		ret = pkvm_create_hyp_vm(kvm);
>  		if (ret)
>  			return ret;
> +
> +		ret = pkvm_create_hyp_vcpu(vcpu);
> +		if (ret)
> +			return ret;
>  	}
>  
>  	mutex_lock(&kvm->arch.config_lock);
> diff --git a/arch/arm64/kvm/hyp/include/nvhe/pkvm.h b/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
> index e42bf68c8848..ce31d3b73603 100644
> --- a/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
> +++ b/arch/arm64/kvm/hyp/include/nvhe/pkvm.h
> @@ -43,12 +43,6 @@ struct pkvm_hyp_vm {
>  	struct hyp_pool pool;
>  	hyp_spinlock_t lock;
>  
> -	/*
> -	 * The number of vcpus initialized and ready to run.
> -	 * Modifying this is protected by 'vm_table_lock'.
> -	 */
> -	unsigned int nr_vcpus;
> -
>  	/* Array of the hyp vCPU structures for this VM. */
>  	struct pkvm_hyp_vcpu *vcpus[];
>  };
> diff --git a/arch/arm64/kvm/hyp/nvhe/pkvm.c b/arch/arm64/kvm/hyp/nvhe/pkvm.c
> index 6efb9bf56180..4ef3748dc660 100644
> --- a/arch/arm64/kvm/hyp/nvhe/pkvm.c
> +++ b/arch/arm64/kvm/hyp/nvhe/pkvm.c
> @@ -245,10 +245,12 @@ struct pkvm_hyp_vcpu *pkvm_load_hyp_vcpu(pkvm_handle_t handle,
>  
>  	hyp_spin_lock(&vm_table_lock);
>  	hyp_vm = get_vm_by_handle(handle);
> -	if (!hyp_vm || hyp_vm->nr_vcpus <= vcpu_idx)
> +	if (!hyp_vm || hyp_vm->kvm.created_vcpus <= vcpu_idx)

Why is it reasonable to trust kvm.created_vcpus here? It looks like
something that is under direct control of the host, and that could be
arbitrarily changed to allow dereferencing the vcpus[] array past its
actual boundaries.

What guarantees that it is stable at load-time?

I have similar misgivings about the init and teardown paths.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.