Re: [PATCH v5 08/10] KVM: x86/MMU: Allow NX huge pages to be disabled on a per-vm basis

[Sean Christopherson]

On Wed, Apr 13, 2022, Ben Gardon wrote:
> diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst
> index 72183ae628f7..021452a9fa91 100644
> --- a/Documentation/virt/kvm/api.rst
> +++ b/Documentation/virt/kvm/api.rst
> @@ -7855,6 +7855,19 @@ At this time, KVM_PMU_CAP_DISABLE is the only capability.  Setting
>  this capability will disable PMU virtualization for that VM.  Usermode
>  should adjust CPUID leaf 0xA to reflect that the PMU is disabled.
>  
> +8.36 KVM_CAP_VM_DISABLE_NX_HUGE_PAGES
> +---------------------------
> +
> +:Capability KVM_CAP_PMU_CAPABILITY
> +:Architectures: x86
> +:Type: vm
> +:Returns 0 on success, -EPERM if the userspace process does not
> +	 have CAP_SYS_BOOT

Needs to document the -EINVAL cases, especially the requirement that this be
called before VMs are created.  The 

> +This capability disables the NX huge pages mitigation for iTLB MULTIHIT.
> +
> +The capability has no effect if the nx_huge_pages module parameter is not set.
> +
>  9. Known KVM API problems
>  =========================
>  
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index 2c20f715f009..b8ab4fa7d4b2 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -1240,6 +1240,8 @@ struct kvm_arch {
>  	hpa_t	hv_root_tdp;
>  	spinlock_t hv_root_tdp_lock;
>  #endif
> +
> +	bool disable_nx_huge_pages;
>  };
>  
>  struct kvm_vm_stat {
> diff --git a/arch/x86/kvm/mmu.h b/arch/x86/kvm/mmu.h
> index 671cfeccf04e..148f630af78a 100644
> --- a/arch/x86/kvm/mmu.h
> +++ b/arch/x86/kvm/mmu.h
> @@ -173,9 +173,10 @@ struct kvm_page_fault {
>  int kvm_tdp_page_fault(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault);
>  
>  extern int nx_huge_pages;
> -static inline bool is_nx_huge_page_enabled(void)
> +static inline bool is_nx_huge_page_enabled(struct kvm *kvm)
>  {
> -	return READ_ONCE(nx_huge_pages);
> +	return READ_ONCE(nx_huge_pages) &&
> +	       !kvm->arch.disable_nx_huge_pages;

No need for a newline, that fits on a single line.

> diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
> index 566548a3efa7..03aa1e0f60e2 100644
> --- a/arch/x86/kvm/mmu/tdp_mmu.c
> +++ b/arch/x86/kvm/mmu/tdp_mmu.c
> @@ -1469,7 +1469,8 @@ static int tdp_mmu_split_huge_page(struct kvm *kvm, struct tdp_iter *iter,
>  	 * not been linked in yet and thus is not reachable from any other CPU.
>  	 */
>  	for (i = 0; i < PT64_ENT_PER_PAGE; i++)
> -		sp->spt[i] = make_huge_page_split_spte(huge_spte, level, i);
> +		sp->spt[i] = make_huge_page_split_spte(kvm, huge_spte,
> +						       level, i);

Just let this poke past 80 chars.

> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 665c1fa8bb57..27631c3b53c2 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -4286,6 +4286,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
>  	case KVM_CAP_SYS_ATTRIBUTES:
>  	case KVM_CAP_VAPIC:
>  	case KVM_CAP_ENABLE_CAP:
> +	case KVM_CAP_VM_DISABLE_NX_HUGE_PAGES:
>  		r = 1;
>  		break;
>  	case KVM_CAP_EXIT_HYPERCALL:
> @@ -6079,6 +6080,28 @@ int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
>  		}
>  		mutex_unlock(&kvm->lock);
>  		break;
> +	case KVM_CAP_VM_DISABLE_NX_HUGE_PAGES:
> +		r = -EINVAL;
> +		if (cap->args[0])
> +			break;
> +
> +		/*
> +		 * Since the risk of disabling NX hugepages is a guest crashing
> +		 * the system, ensure the userspace process has permission to
> +		 * reboot the system.

Since I'm nitpicking already and there's also a comment...

Can you call out that, unlike the actual reboot() syscall, the process needs the
capability in the init? namespace (I don't actual know the terminology) because
exposing /dev/kvm into a container doesn't magically limit the iTLB multihit bug
to that container.  I.e. that this _must_ use capable(), not ns_capable().	

Amusingly, someone could subvert the selftest's SYS_reboot heuristic by running
the test in a container :-)