Re: [PATCH v21 05/23] KVM: arm64: Track the type of VM in kvm_arch

[Suzuki K Poulose]

On 02/10/2026 15:15, Marc Zyngier wrote:
> On Fri, 02 Oct 2026 14:05:17 +0100,
> Fuad Tabba <tabba@xxxxxxxxxx> wrote:
> > Hi Marc,
> >
> > On Fri, 02 Oct 2026 13:57:11 +0100, Marc Zyngier <maz@xxxxxxxxxx> wrote:
> > [...]
> > > Just to make sure. Is that what you had in mind?
> >
> > A diff is worth a thousand words :) Yes, that's it. With that, every
> > remaining hyp user of vcpu_is_protected() only runs with pKVM, so the
> > kern_hyp_va() case in the macro can go too.
>
> Actually, scratch all of that. The whole thing is written upside down,
> and makes little sense. I wish I had looked at it more closely...
>
> The *_protected__pkvm() shouldn't exist at all, because nVHE is the
> only context where it matters, and therefore we can specialise the
> *normal* accessors for the nVHE context (see below). No *_pkvm(), no
> kern_hyp_va(), nothing at all.
>
> I'm also dead against the unprotected_pkvm() stuff, as it looks
> pointlessly specialised. Specifically, this nugget is making my head
> spin:
>
> +       if (!kvm_vm_is_unprotected_pkvm(vcpu->kvm))
>                  return;
>
> Honestly, we introduce the flavor stuff to make it easy to match
> things on a particular VM type in a readable way. So why isn't this
> reading:
>
> 	if (vcpu->kvm->arch.vm_flavor != VM_PROTECTED_PKVM)
> 		return;
>
> which makes it pretty explicit. You can gate it with a
> is_protected_kvm_enabled() if you want, as it was before.
>
> 	M.
>
>
> diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> index 9e1fa00d96f7c..6b64f9610f3a3 100644
> --- a/arch/arm64/include/asm/kvm_host.h
> +++ b/arch/arm64/include/asm/kvm_host.h
> @@ -1519,41 +1519,22 @@ struct kvm *kvm_arch_alloc_vm(void);
>   
>   #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE
>   
> -#define kvm_vm_is_protected(kvm)	((kvm)->arch.vm_flavor >= __VM_PROTECTED)
> -/*
> - * Accessing vcpu->kvm from nVHE hyp stub is tricky, as we need to convert the
> - * pointer to the hyp VA. With pKVM, the nVHE code runs with the hyp_vcpu,
> - * which is populated correctly.
> - */
> -#define vcpu_is_protected(vcpu)						\
> -	({								\
> -		struct kvm *__kvm = READ_ONCE((vcpu)->kvm);		\
> -		bool __protected = false;				\
> -									\
> -		if (__kvm) {						\
> -			if (is_nvhe_hyp_code() &&			\
> -			    !is_protected_kvm_enabled()) 		\
> -				__kvm = kern_hyp_va(__kvm); 		\
> -									\
> -			__protected = kvm_vm_is_protected(__kvm);	\
> -		}							\
> -		__protected;						\
> -	})
> -
> -#define kvm_vm_is_protected_pkvm(kvm)					\
> +#ifdef __KVM_NVHE_HYPERVISOR__
> +#define kvm_vm_is_protected(kvm)					\
>   	(is_protected_kvm_enabled() && ((kvm)->arch.vm_flavor == VM_PROTECTED_PKVM))
> -
>   /*
>    * Rely on is_protected_kvm_enabled() check in kvm_vm_is_protected_pkvm() to
>    * make sure the vcpu->kvm is always valid VA in the context
>    */
> -#define vcpu_is_protected_pkvm(vcpu)					\
> +#define vcpu_is_protected(vcpu)						\
>   	({								\
>   		struct kvm *__kvm = READ_ONCE((vcpu)->kvm);		\
> -									\
> -		(__kvm && kvm_vm_is_protected_pkvm(__kvm));		\
> +		(__kvm && kvm_vm_is_protected(__kvm));			\
>   	})
> -
> +#else
> +#define kvm_vm_is_protected(kvm)	((kvm)->arch.vm_flavor >= __VM_PROTECTED)
> +#define vcpu_is_protected(vcpu)		kvm_vm_is_protected((vcpu)->kvm)
> +#endif
>   
>   #define kvm_vm_is_unprotected_pkvm(kvm)		\
>   	(is_protected_kvm_enabled() && ((kvm)->arch.vm_flavor == VM_PKVM))
> diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
> index 0f4e8b71fa85d..9ba86450fe4af 100644
> --- a/arch/arm64/kvm/mmu.c
> +++ b/arch/arm64/kvm/mmu.c
> @@ -2624,7 +2624,7 @@ int kvm_arch_prepare_memory_region(struct kvm *kvm,
>   	hva_t hva, reg_end;
>   	int ret = 0;
>   
> -	if (kvm_vm_is_protected_pkvm(kvm)) {
> +	if (kvm_vm_is_protected(kvm)) {

We have slightly different handling for Realms. The final code for this
looks like :

	if (kvm_vm_is_protected(kvm)) {
		if (new &&
		    new->flags & (KVM_MEM_LOG_DIRTY_PAGES | KVM_MEM_READONLY)) {
			return -EPERM;
		}
	}

	if (kvm_vm_is_protected_pkvm(kvm)) {
		/* Cannot modify memslots once a pVM has run. */
		if (pkvm_hyp_vm_is_created(kvm) &&
		    (change == KVM_MR_DELETE || change == KVM_MR_MOVE)) {
			return -EPERM;
		}
	} else if (kvm_realm_is_created(kvm)) {
		/*
		 * Once the Realm is created, we cannot modify any slots that
		 * could be providing private memory. i.e., guest_memfd backed
		 * slots.
		 * TODO: Handle trusted device private memory slots
		 */
		if (kvm_slot_has_gmem(old) || kvm_slot_has_gmem(new))
			return -EPERM;
	}


Of course we could invert the checks to detect vm_is_protected_pkvm()

Suzuki

>   		/* Cannot modify memslots once a pVM has run. */
>   		if (pkvm_hyp_vm_is_created(kvm) &&
>   		    (change == KVM_MR_DELETE || change == KVM_MR_MOVE)) {