[PATCH v2 4/8] KVM: SEV: Add direct VMSA capability

[Jörg Rödel]

From: Joerg Roedel <joerg.roedel@xxxxxxx>

Direct VMSA mode is currently selected implicitly when userspace installs
a guest-owned VMSA on any vCPU.  That makes the launch model a side effect
of per-vCPU state and applies it retroactively to every vCPU.

Add KVM_CAP_SNP_DIRECT_VMSA as a VM-scoped settable capability.  Require
userspace to enable it on an SNP VM before creating vCPUs, and reject
nonzero arguments.

When enabled, skip allocating KVM-owned VMSAs during vCPU creation and
skip generating and measuring KVM-owned VMSAs at launch finish.  Keep
legacy allocation and measurement behavior when it is not enabled.

Assisted-by: LLM
---
 arch/x86/include/asm/kvm-x86-ops.h |  1 +
 arch/x86/include/asm/kvm_host.h    |  1 +
 arch/x86/kvm/svm/sev.c             | 28 ++++++++++++++++++++--------
 arch/x86/kvm/svm/svm.c             | 22 ++++++++++++++++++++++
 arch/x86/kvm/svm/svm.h             |  1 +
 arch/x86/kvm/x86.c                 |  5 +++++
 include/uapi/linux/kvm.h           |  1 +
 7 files changed, 51 insertions(+), 8 deletions(-)

diff --git a/arch/x86/include/asm/kvm-x86-ops.h b/arch/x86/include/asm/kvm-x86-ops.h
index e213c9ae3e30..a48259696ca6 100644
--- a/arch/x86/include/asm/kvm-x86-ops.h
+++ b/arch/x86/include/asm/kvm-x86-ops.h
@@ -23,6 +23,7 @@ KVM_X86_OP(vcpu_after_set_cpuid)
 KVM_X86_OP(vm_init)
 KVM_X86_OP_OPTIONAL(vm_destroy)
 KVM_X86_OP_OPTIONAL(vm_pre_destroy)
+KVM_X86_OP_OPTIONAL(enable_vm_cap)
 KVM_X86_OP_OPTIONAL_RET0(vcpu_precreate)
 KVM_X86_OP(vcpu_create)
 KVM_X86_OP(vcpu_free)
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 683bb8bf43a9..3bfc02bb7998 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -1519,6 +1519,7 @@ struct kvm_x86_ops {
 	int (*vm_init)(struct kvm *kvm);
 	void (*vm_destroy)(struct kvm *kvm);
 	void (*vm_pre_destroy)(struct kvm *kvm);
+	int (*enable_vm_cap)(struct kvm *kvm, struct kvm_enable_cap *cap);
 
 	/* Create, but do not attach this VCPU */
 	int (*vcpu_precreate)(struct kvm *kvm);
diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
index ee1bf9b4cc53..5a282aff04a8 100644
--- a/arch/x86/kvm/svm/sev.c
+++ b/arch/x86/kvm/svm/sev.c
@@ -2524,8 +2524,15 @@ static int snp_launch_update_vmsa(struct kvm *kvm, struct kvm_sev_cmd *argp)
 
 	kvm_for_each_vcpu(i, vcpu, kvm) {
 		struct vcpu_svm *svm = to_svm(vcpu);
-		u64 pfn = __pa(svm->sev_es.vmsa) >> PAGE_SHIFT;
+		u64 pfn;
+
+		if (sev->snp_direct_vmsa) {
+			if (!svm->sev_es.snp_has_guest_vmsa)
+				svm->vmcb->control.vmsa_pa = INVALID_PAGE;
+			goto protect_vcpu;
+		}
 
+		pfn = __pa(svm->sev_es.vmsa) >> PAGE_SHIFT;
 		ret = sev_es_sync_vmsa(svm);
 		if (ret)
 			goto out;
@@ -2545,6 +2552,7 @@ static int snp_launch_update_vmsa(struct kvm *kvm, struct kvm_sev_cmd *argp)
 			goto out;
 		}
 
+protect_vcpu:
 		svm->vcpu.arch.guest_state_protected = true;
 		/*
 		 * SEV-ES (and thus SNP) guest mandates LBR Virtualization to
@@ -3559,6 +3567,9 @@ void sev_free_vcpu(struct kvm_vcpu *vcpu)
 	 * a guest-owned page. Transition the page to hypervisor state before
 	 * releasing it back to the system.
 	 */
+	if (!svm->sev_es.vmsa)
+		goto skip_vmsa_free;
+
 	if (is_sev_snp_guest(vcpu)) {
 		u64 pfn = __pa(svm->sev_es.vmsa) >> PAGE_SHIFT;
 
@@ -4056,13 +4067,9 @@ static void __sev_snp_reload_vmsa(struct kvm_vcpu *vcpu, gpa_t gpa)
 	vmcb_mark_all_dirty(svm->vmcb);
 
 	/*
-	 * From this point forward, the VMSA will always be a guest-mapped page
-	 * rather than the initial one allocated by KVM in svm->sev_es.vmsa. In
-	 * theory, svm->sev_es.vmsa could be free'd and cleaned up here, but
-	 * that involves cleanups like flushing caches, which would ideally be
-	 * handled during teardown rather than guest boot.  Deferring that also
-	 * allows the existing logic for SEV-ES VMSAs to be re-used with
-	 * minimal SNP-specific changes.
+	 * From this point forward, the VMSA will always be a guest-mapped page.
+	 * If KVM allocated an initial VMSA, keep it until teardown to defer
+	 * cache flushing and other cleanup out of the guest boot path.
 	 */
 	svm->sev_es.snp_has_guest_vmsa = true;
 
@@ -4929,12 +4936,15 @@ void sev_init_vmcb(struct vcpu_svm *svm, bool init_event)
 int sev_vcpu_create(struct kvm_vcpu *vcpu)
 {
 	struct vcpu_svm *svm = to_svm(vcpu);
+	struct kvm_sev_info *sev = to_kvm_sev_info(vcpu->kvm);
 	struct page *vmsa_page;
 
 	mutex_init(&svm->sev_es.snp_vmsa_mutex);
 
 	if (!is_sev_es_guest(vcpu))
 		return 0;
+	if (is_sev_snp_guest(vcpu) && sev->snp_direct_vmsa)
+		goto init_vmsa_state;
 
 	/*
 	 * SEV-ES guests require a separate (from the VMCB) VMSA page used to
@@ -4945,6 +4955,8 @@ int sev_vcpu_create(struct kvm_vcpu *vcpu)
 		return -ENOMEM;
 
 	svm->sev_es.vmsa = page_address(vmsa_page);
+
+init_vmsa_state:
 	svm->sev_es.snp_pending_vmsa_gpa = INVALID_PAGE;
 	svm->sev_es.snp_guest_vmsa_gpa = INVALID_PAGE;
 
diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
index 7d59d301e1e5..98c5cffe2c22 100644
--- a/arch/x86/kvm/svm/svm.c
+++ b/arch/x86/kvm/svm/svm.c
@@ -5325,6 +5325,27 @@ static void *svm_alloc_apic_backing_page(struct kvm_vcpu *vcpu)
 	return page_address(page);
 }
 
+static int svm_enable_vm_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
+{
+	switch (cap->cap) {
+#ifdef CONFIG_KVM_AMD_SEV
+	case KVM_CAP_SNP_DIRECT_VMSA:
+		if (memchr_inv(cap->args, 0, sizeof(cap->args)) ||
+		    kvm->arch.vm_type != KVM_X86_SNP_VM)
+			return -EINVAL;
+
+		guard(mutex)(&kvm->lock);
+		if (kvm->created_vcpus)
+			return -EINVAL;
+
+		to_kvm_sev_info(kvm)->snp_direct_vmsa = true;
+		return 0;
+#endif
+	default:
+		return -EINVAL;
+	}
+}
+
 struct kvm_x86_ops svm_x86_ops __initdata = {
 	.name = KBUILD_MODNAME,
 
@@ -5345,6 +5366,7 @@ struct kvm_x86_ops svm_x86_ops __initdata = {
 	.vm_init = svm_vm_init,
 	.vm_pre_destroy = avic_vm_pre_destroy,
 	.vm_destroy = svm_vm_destroy,
+	.enable_vm_cap = svm_enable_vm_cap,
 
 	.prepare_switch_to_guest = svm_prepare_switch_to_guest,
 	.vcpu_load = svm_vcpu_load,
diff --git a/arch/x86/kvm/svm/svm.h b/arch/x86/kvm/svm/svm.h
index e958943b8162..881d662aaab2 100644
--- a/arch/x86/kvm/svm/svm.h
+++ b/arch/x86/kvm/svm/svm.h
@@ -121,6 +121,7 @@ struct kvm_sev_info {
 	struct mutex guest_req_mutex; /* Must acquire before using bounce buffers */
 	cpumask_var_t have_run_cpus; /* CPUs that have done VMRUN for this VM. */
 	bool snp_certs_enabled;	/* SNP certificate-fetching support. */
+	bool snp_direct_vmsa;	/* Userspace provides and measures VMSA pages. */
 };
 #endif
 
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 79468ddfe473..fb4857eca6c6 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -2404,6 +2404,9 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 	case KVM_CAP_VM_TYPES:
 		r = kvm_caps.supported_vm_types;
 		break;
+	case KVM_CAP_SNP_DIRECT_VMSA:
+		r = !!(kvm_caps.supported_vm_types & BIT(KVM_X86_SNP_VM));
+		break;
 	case KVM_CAP_READONLY_MEM:
 		r = kvm ? kvm_arch_has_readonly_mem(kvm) : 1;
 		break;
@@ -4212,6 +4215,8 @@ int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
 	}
 	default:
 		r = -EINVAL;
+		if (kvm_x86_ops.enable_vm_cap)
+			r = kvm_x86_call(enable_vm_cap)(kvm, cap);
 		break;
 	}
 	return r;
diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
index ac2d77d14963..283e881b2634 100644
--- a/include/uapi/linux/kvm.h
+++ b/include/uapi/linux/kvm.h
@@ -999,6 +999,7 @@ struct kvm_enable_cap {
 #define KVM_CAP_S390_HPAGE_2G 249
 #define KVM_CAP_PPC_COMPAT_CAPS 250
 #define KVM_CAP_ARM_PMU_V3_STRICT 251
+#define KVM_CAP_SNP_DIRECT_VMSA 252
 
 struct kvm_irq_routing_irqchip {
 	__u32 irqchip;
-- 
2.53.0