linux-kernel - [PATCH 21/29] KVM: x86: add infrastructure to share FPU across planes

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20250401161106.790710-22-pbonzini@redhat.com>
Date: Tue,  1 Apr 2025 18:10:58 +0200
From: Paolo Bonzini <pbonzini@...hat.com>
To: linux-kernel@...r.kernel.org,
	kvm@...r.kernel.org
Cc: roy.hopkins@...e.com,
	seanjc@...gle.com,
	thomas.lendacky@....com,
	ashish.kalra@....com,
	michael.roth@....com,
	jroedel@...e.de,
	nsaenz@...zon.com,
	anelkz@...zon.de,
	James.Bottomley@...senPartnership.com
Subject: [PATCH 21/29] KVM: x86: add infrastructure to share FPU across planes

Wrap fpu_alloc_guest_fpstate() and fpu_free_guest_fpstate() so that only
one FPU exists for vCPUs that are in different planes but share the same
vCPU id.

This API could be handy for VTL implementation but it may be tricky
because for some registers sharing would be a bad idea (even MPX right
now if it weren't deprecated, but APX in the future could be worse).

Signed-off-by: Paolo Bonzini <pbonzini@...hat.com>
---
 arch/x86/include/asm/kvm_host.h |  3 +++
 arch/x86/kvm/x86.c              | 47 ++++++++++++++++++++++++++++-----
 2 files changed, 44 insertions(+), 6 deletions(-)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 283d8a4b5b14..9ac39f128a53 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -1347,6 +1347,7 @@ struct kvm_arch {
 	unsigned int indirect_shadow_pages;
 	u8 mmu_valid_gen;
 	u8 vm_type;
+	bool planes_share_fpu;
 	bool has_private_mem;
 	bool has_protected_state;
 	bool pre_fault_allowed;
@@ -2447,4 +2448,6 @@ int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages);
  */
 #define KVM_EXIT_HYPERCALL_MBZ		GENMASK_ULL(31, 1)
 
+bool kvm_arch_planes_share_fpu(struct kvm *kvm);
+
 #endif /* _ASM_X86_KVM_HOST_H */
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index ce8e623052a7..ebdbd08a840b 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -6626,6 +6626,17 @@ int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
 		kvm->arch.triple_fault_event = cap->args[0];
 		r = 0;
 		break;
+	case KVM_CAP_PLANES_FPU:
+		r = -EINVAL;
+		if (atomic_read(&kvm->online_vcpus))
+			break;
+		if (cap->args[0] > 1)
+			break;
+		if (cap->args[0] && kvm->arch.has_protected_state)
+			break;
+		kvm->arch.planes_share_fpu = cap->args[0];
+		r = 0;
+		break;
 	case KVM_CAP_X86_USER_SPACE_MSR:
 		r = -EINVAL;
 		if (cap->args[0] & ~KVM_MSR_EXIT_REASON_VALID_MASK)
@@ -12332,6 +12343,27 @@ int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
 	return kvm_x86_call(vcpu_precreate)(kvm);
 }
 
+static void kvm_free_guest_fpstate(struct kvm_vcpu *vcpu, unsigned plane)
+{
+	if (plane == 0 || !vcpu->kvm->arch.planes_share_fpu)
+		fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
+}
+
+static int kvm_init_guest_fpstate(struct kvm_vcpu *vcpu, struct kvm_vcpu *plane0_vcpu)
+{
+	if (plane0_vcpu && vcpu->kvm->arch.planes_share_fpu) {
+		vcpu->arch.guest_fpu = plane0_vcpu->arch.guest_fpu;
+		return 0;
+	}
+
+	if (!fpu_alloc_guest_fpstate(&vcpu->arch.guest_fpu)) {
+		pr_err("failed to allocate vcpu's fpu\n");
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+
 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu, struct kvm_plane *plane)
 {
 	struct page *page;
@@ -12378,10 +12410,8 @@ int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu, struct kvm_plane *plane)
 	if (!alloc_emulate_ctxt(vcpu))
 		goto free_wbinvd_dirty_mask;
 
-	if (!fpu_alloc_guest_fpstate(&vcpu->arch.guest_fpu)) {
-		pr_err("failed to allocate vcpu's fpu\n");
+	if (kvm_init_guest_fpstate(vcpu, plane->plane ? vcpu->plane0 : NULL) < 0)
 		goto free_emulate_ctxt;
-	}
 
 	kvm_async_pf_hash_reset(vcpu);
 
@@ -12413,7 +12443,7 @@ int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu, struct kvm_plane *plane)
 	return 0;
 
 free_guest_fpu:
-	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
+	kvm_free_guest_fpstate(vcpu, plane->plane);
 free_emulate_ctxt:
 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
 free_wbinvd_dirty_mask:
@@ -12459,7 +12489,7 @@ void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 
 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
-	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
+	kvm_free_guest_fpstate(vcpu, vcpu->plane);
 
 	kvm_xen_destroy_vcpu(vcpu);
 	kvm_hv_vcpu_uninit(vcpu);
@@ -12824,7 +12854,7 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 	kvm->arch.apic_bus_cycle_ns = APIC_BUS_CYCLE_NS_DEFAULT;
 	kvm->arch.guest_can_read_msr_platform_info = true;
 	kvm->arch.enable_pmu = enable_pmu;
-
+	kvm->arch.planes_share_fpu = false;
 #if IS_ENABLED(CONFIG_HYPERV)
 	spin_lock_init(&kvm->arch.hv_root_tdp_lock);
 	kvm->arch.hv_root_tdp = INVALID_PAGE;
@@ -13881,6 +13911,11 @@ int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
 }
 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
 
+bool kvm_arch_planes_share_fpu(struct kvm *kvm)
+{
+	return !kvm || kvm->arch.planes_share_fpu;
+}
+
 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
 {
 	struct kvm_run *run = vcpu->run;
-- 
2.49.0