Re: [PATCH v2 09/11] KVM: x86/MMU: Allow NX huge pages to be disabled on a per-vm basis

[David Matlack <dmatlack@google.com>]

On Mon, Mar 21, 2022 at 04:48:42PM -0700, Ben Gardon wrote:
> In some cases, the NX hugepage mitigation for iTLB multihit is not
> needed for all guests on a host. Allow disabling the mitigation on a
> per-VM basis to avoid the performance hit of NX hugepages on trusted
> workloads.
> 
> Signed-off-by: Ben Gardon <bgardon@google.com>
> ---
>  arch/x86/include/asm/kvm_host.h | 1 +
>  arch/x86/kvm/mmu.h              | 1 +
>  arch/x86/kvm/mmu/mmu.c          | 6 ++++--
>  arch/x86/kvm/x86.c              | 6 ++++++
>  include/uapi/linux/kvm.h        | 1 +
>  5 files changed, 13 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index 0a0c54639dd8..04ddfc475ce0 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -1242,6 +1242,7 @@ struct kvm_arch {
>  #endif
>  
>  	bool nx_huge_pages;
> +	bool disable_nx_huge_pages;
>  };
>  
>  struct kvm_vm_stat {
> diff --git a/arch/x86/kvm/mmu.h b/arch/x86/kvm/mmu.h
> index dd28fe8d13ae..36d8d84ca6c6 100644
> --- a/arch/x86/kvm/mmu.h
> +++ b/arch/x86/kvm/mmu.h
> @@ -177,6 +177,7 @@ static inline bool is_nx_huge_page_enabled(struct kvm *kvm)
>  {
>  	return READ_ONCE(kvm->arch.nx_huge_pages);
>  }
> +void kvm_update_nx_huge_pages(struct kvm *kvm);
>  
>  static inline int kvm_mmu_do_page_fault(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
>  					u32 err, bool prefetch)
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> index dc9672f70468..a7d387ccfd74 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -6195,9 +6195,10 @@ static void __set_nx_huge_pages(bool val)
>  	nx_huge_pages = itlb_multihit_kvm_mitigation = val;
>  }
>  
> -static void kvm_update_nx_huge_pages(struct kvm *kvm)
> +void kvm_update_nx_huge_pages(struct kvm *kvm)
>  {
> -	kvm->arch.nx_huge_pages = nx_huge_pages;
> +	kvm->arch.nx_huge_pages = nx_huge_pages &&
> +				  !kvm->arch.disable_nx_huge_pages;

kvm->arch.nx_huge_pages seems like it could be dropped and
is_nx_huge_page_enabled() could just check this condition.

>  
>  	mutex_lock(&kvm->slots_lock);
>  	kvm_mmu_zap_all_fast(kvm);
> @@ -6451,6 +6452,7 @@ int kvm_mmu_post_init_vm(struct kvm *kvm)
>  	int err;
>  
>  	kvm->arch.nx_huge_pages = READ_ONCE(nx_huge_pages);
> +	kvm->arch.disable_nx_huge_pages = false;

I believe this can be omitted since kvm_arch is zero-initialized.

>  	err = kvm_vm_create_worker_thread(kvm, kvm_nx_lpage_recovery_worker, 0,
>  					  "kvm-nx-lpage-recovery",
>  					  &kvm->arch.nx_lpage_recovery_thread);
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 51106d32f04e..73df90a6932b 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -4256,6 +4256,7 @@ 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:

Please document the new capability.

>  		r = 1;
>  		break;
>  	case KVM_CAP_EXIT_HYPERCALL:
> @@ -6048,6 +6049,11 @@ int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
>  		}
>  		mutex_unlock(&kvm->lock);
>  		break;
> +	case KVM_CAP_VM_DISABLE_NX_HUGE_PAGES:
> +		kvm->arch.disable_nx_huge_pages = true;
> +		kvm_update_nx_huge_pages(kvm);
> +		r = 0;
> +		break;
>  	default:
>  		r = -EINVAL;
>  		break;
> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> index ee5cc9e2a837..6f9fa7ecfd1e 100644
> --- a/include/uapi/linux/kvm.h
> +++ b/include/uapi/linux/kvm.h
> @@ -1144,6 +1144,7 @@ struct kvm_ppc_resize_hpt {
>  #define KVM_CAP_S390_MEM_OP_EXTENSION 211
>  #define KVM_CAP_PMU_CAPABILITY 212
>  #define KVM_CAP_DISABLE_QUIRKS2 213
> +#define KVM_CAP_VM_DISABLE_NX_HUGE_PAGES 214
>  
>  #ifdef KVM_CAP_IRQ_ROUTING
>  
> -- 
> 2.35.1.894.gb6a874cedc-goog
>