Skip to contentSkip to conversation

KVM: x86: Provide per VM capability for disabling PMU virtualization

PATCH Superseded David Dunn 3 participants 7 messages Jan 23, 2022 – Feb 9, 2022

Subsystems (5)
Documentation x86 KVM Kselftest +1 more subsystems

v5 of 7 Superseded Latest is v7 →

lore
Share & tools
lore

David Dunn

author
#0 lore raw

v5 cover KVM: x86: Provide per VM capability for disabling PMU virtualization

Changes since v4:
- Remove automatic CPUID adjustment when PMU disabled [Like]
- Update documentation and changelog to reflect above.
- Update documentation to document arg[0] and return values. [Like].

Original description:

This patch set allows usermode to disable PMU virtualization on individual x86 VMs. When disabled, the PMU is not advertised to or accessible from the guest.

David Dunn (3):
KVM: x86: Provide per VM capability for disabling PMU virtualization
KVM: selftests: Allow creation of selftest VM without vcpus
KVM: selftests: Verify disabling PMU virtualization via
KVM_CAP_CONFIG_PMU

 Documentation/virt/kvm/api.rst                | 22 +++++++++
 arch/x86/include/asm/kvm_host.h               |  1 +
 arch/x86/kvm/svm/pmu.c                        |  2 +-
 arch/x86/kvm/vmx/pmu_intel.c                  |  2 +-
 arch/x86/kvm/x86.c                            | 12 +++++
 include/uapi/linux/kvm.h                      |  4 ++
 tools/include/uapi/linux/kvm.h                |  4 ++
 .../selftests/kvm/include/kvm_util_base.h     |  3 ++
 tools/testing/selftests/kvm/lib/kvm_util.c    | 48 ++++++++++++++-----
 .../kvm/x86_64/pmu_event_filter_test.c        | 35 ++++++++++++++
 10 files changed, 118 insertions(+), 15 deletions(-)
Signature · 1 line
2.35.0.rc0.227.g00780c9af4-goog
author
#1 lore raw

PATCH v5 1/3 KVM: x86: Provide per VM capability for disabling PMU virtualization

KVM_CAP_PMU_DISABLE is used to disable PMU virtualization on individual x86 VMs. PMU configuration must be done prior to creating VCPUs.

To enable future extension, KVM_CAP_PMU_CAPABILITY reports available settings via bitmask when queried via check_extension.

For VMs that have PMU virtualization disabled, usermode will need to clear CPUID leaf 0xA to notify guests.

---
 Documentation/virt/kvm/api.rst  | 22 ++++++++++++++++++++++
 arch/x86/include/asm/kvm_host.h |  1 +
 arch/x86/kvm/svm/pmu.c          |  2 +-
 arch/x86/kvm/vmx/pmu_intel.c    |  2 +-
 arch/x86/kvm/x86.c              | 12 ++++++++++++
 include/uapi/linux/kvm.h        |  4 ++++
 tools/include/uapi/linux/kvm.h  |  4 ++++
 7 files changed, 45 insertions(+), 2 deletions(-)
Documentation/virt/kvm/api.rst +22 -0
index bb8cfddbb22d..ae0d52c5359e 100644
@@ -7559,3 +7559,25 @@ The argument to KVM_ENABLE_CAP is also a bitmask, and must be a subset
 of the result of KVM_CHECK_EXTENSION.  KVM will forward to userspace
 the hypercalls whose corresponding bit is in the argument, and return
 ENOSYS for the others.
+
+8.35 KVM_CAP_PMU_CAPABILITY
+---------------------------
+
+:Capability KVM_CAP_PMU_CAPABILITY
+:Architectures: x86
+:Type: vm
+:Parameters: arg[0] is bitmask of PMU virtualization capabilities.
+:Returns 0 on success, -EINVAL when arg[0] contains invalid bits
+
+This capability alters PMU virtualization in KVM.
+
+Calling KVM_CHECK_EXTENSION for this capability returns a bitmask of
+PMU virtualization capabilities that can be adjusted on a VM.
+
+The argument to KVM_ENABLE_CAP is also a bitmask and selects specific
+PMU virtualization capabilities to be applied to the VM.  This can
+only be invoked on a VM prior to the creation of VCPUs.
+
+At this time, KVM_CAP_PMU_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.
arch/x86/include/asm/kvm_host.h +1 -0
index 682ad02a4e58..5cdcd4a7671b 100644
@@ -1232,6 +1232,7 @@ struct kvm_arch {
 	hpa_t	hv_root_tdp;
 	spinlock_t hv_root_tdp_lock;
 #endif
+	bool enable_pmu;
 };
 
 struct kvm_vm_stat {
arch/x86/kvm/svm/pmu.c +1 -1
index 5aa45f13b16d..d4de52409335 100644
@@ -101,7 +101,7 @@ static inline struct kvm_pmc *get_gp_pmc_amd(struct kvm_pmu *pmu, u32 msr,
 {
 	struct kvm_vcpu *vcpu = pmu_to_vcpu(pmu);
 
-	if (!enable_pmu)
+	if (!vcpu->kvm->arch.enable_pmu)
 		return NULL;
 
 	switch (msr) {
arch/x86/kvm/vmx/pmu_intel.c +1 -1
index 466d18fc0c5d..2c5868d77268 100644
@@ -487,7 +487,7 @@ static void intel_pmu_refresh(struct kvm_vcpu *vcpu)
 	pmu->reserved_bits = 0xffffffff00200000ull;
 
 	entry = kvm_find_cpuid_entry(vcpu, 0xa, 0);
-	if (!entry || !enable_pmu)
+	if (!entry || !vcpu->kvm->arch.enable_pmu)
 		return;
 	eax.full = entry->eax;
 	edx.full = entry->edx;
arch/x86/kvm/x86.c +12 -0
index 55518b7d3b96..a033f019a3f0 100644
@@ -4326,6 +4326,9 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 		if (r < sizeof(struct kvm_xsave))
 			r = sizeof(struct kvm_xsave);
 		break;
+	case KVM_CAP_PMU_CAPABILITY:
+		r = enable_pmu ? KVM_CAP_PMU_MASK : 0;
+		break;
 	}
 	default:
 		break;
@@ -5937,6 +5940,14 @@ int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
 		kvm->arch.exit_on_emulation_error = cap->args[0];
 		r = 0;
 		break;
+	case KVM_CAP_PMU_CAPABILITY:
+		r = -EINVAL;
+		if (!enable_pmu || kvm->created_vcpus > 0 ||
+		    cap->args[0] & ~KVM_CAP_PMU_MASK)
+			break;
+		kvm->arch.enable_pmu = !(cap->args[0] & KVM_CAP_PMU_DISABLE);
+		r = 0;
+		break;
 	default:
 		r = -EINVAL;
 		break;
@@ -11562,6 +11573,7 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
 
 	kvm->arch.guest_can_read_msr_platform_info = true;
+	kvm->arch.enable_pmu = enable_pmu;
 
 #if IS_ENABLED(CONFIG_HYPERV)
 	spin_lock_init(&kvm->arch.hv_root_tdp_lock);
include/uapi/linux/kvm.h +4 -0
index 9563d294f181..1c5e6e172817 100644
@@ -1133,6 +1133,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM 206
 #define KVM_CAP_VM_GPA_BITS 207
 #define KVM_CAP_XSAVE2 208
+#define KVM_CAP_PMU_CAPABILITY 209
 
 #ifdef KVM_CAP_IRQ_ROUTING
 
@@ -1972,6 +1973,9 @@ struct kvm_dirty_gfn {
 #define KVM_BUS_LOCK_DETECTION_OFF             (1 << 0)
 #define KVM_BUS_LOCK_DETECTION_EXIT            (1 << 1)
 
+#define KVM_CAP_PMU_DISABLE                    (1 << 0)
+#define KVM_CAP_PMU_MASK                       (KVM_CAP_PMU_DISABLE)
+
 /**
  * struct kvm_stats_header - Header of per vm/vcpu binary statistics data.
  * @flags: Some extra information for header, always 0 for now.
tools/include/uapi/linux/kvm.h +4 -0
index 9563d294f181..a361cf6e8604 100644
@@ -1133,6 +1133,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM 206
 #define KVM_CAP_VM_GPA_BITS 207
 #define KVM_CAP_XSAVE2 208
+#define KVM_CAP_PMU_CAPABILITY 209
 
 #ifdef KVM_CAP_IRQ_ROUTING
 
@@ -1972,6 +1973,9 @@ struct kvm_dirty_gfn {
 #define KVM_BUS_LOCK_DETECTION_OFF             (1 << 0)
 #define KVM_BUS_LOCK_DETECTION_EXIT            (1 << 1)
 
+#define KVM_CAP_PMU_DISABLE                    (1 << 0)
+#define KVM_CAP_PMU_MASK                       (KVM_PMU_CONFIG_DISABLE)
+
 /**
  * struct kvm_stats_header - Header of per vm/vcpu binary statistics data.
  * @flags: Some extra information for header, always 0 for now.
Signature · 1 line
2.35.0.rc0.227.g00780c9af4-goog
Reviewed-by on v5 1/3
#2 lore raw
Quoted 11 lines · David Dunn

KVM_CAP_PMU_DISABLE is used to disable PMU virtualization on individual x86 VMs. PMU configuration must be done prior to creating VCPUs.

To enable future extension, KVM_CAP_PMU_CAPABILITY reports available settings via bitmask when queried via check_extension.

For VMs that have PMU virtualization disabled, usermode will need to clear CPUID leaf 0xA to notify guests.

Nit: The two references to CPUID leaf 0xA should be qualified as applying only to Intel VMs.

Reviewed-by
#3 lore raw

cc LKML and full list of KVM reviewers.

Quoted 11 lines · Jim Mattson
Quoted 10 lines · David Dunn

KVM_CAP_PMU_DISABLE is used to disable PMU virtualization on individual x86 VMs. PMU configuration must be done prior to creating VCPUs.

To enable future extension, KVM_CAP_PMU_CAPABILITY reports available settings via bitmask when queried via check_extension.

For VMs that have PMU virtualization disabled, usermode will need to clear CPUID leaf 0xA to notify guests.

Quoted 7 lines
Quoted 1 line

Nit: The two references to CPUID leaf 0xA should be qualified as applying only to Intel VMs.

Nit: It looks like we already have "#define KVM_CAP_SYS_ATTRIBUTES 209".

Hope it helps a little:

author
#4 lore raw

Thanks Like.

I just sent a new v6 series with these minor merge conflicts resolved.

Dave

Quoted 28 lines · Like Xu

cc LKML and full list of KVM reviewers.

Quoted 11 lines · Jim Mattson
Quoted 10 lines · David Dunn

KVM_CAP_PMU_DISABLE is used to disable PMU virtualization on individual x86 VMs. PMU configuration must be done prior to creating VCPUs.

To enable future extension, KVM_CAP_PMU_CAPABILITY reports available settings via bitmask when queried via check_extension.

For VMs that have PMU virtualization disabled, usermode will need to clear CPUID leaf 0xA to notify guests.

Quoted 7 lines
Quoted 1 line

Nit: The two references to CPUID leaf 0xA should be qualified as applying only to Intel VMs.

Nit: It looks like we already have "#define KVM_CAP_SYS_ATTRIBUTES 209".

Hope it helps a little:

author
#5 lore raw

PATCH v5 2/3 KVM: selftests: Allow creation of selftest VM without vcpus

In reply to David Dunn

Break out portion of vm_create_with_vcpus so that selftests can modify the VM prior to creating vcpus.

---
 .../selftests/kvm/include/kvm_util_base.h     |  3 ++
 tools/testing/selftests/kvm/lib/kvm_util.c    | 48 ++++++++++++++-----
 2 files changed, 38 insertions(+), 13 deletions(-)
tools/testing/selftests/kvm/include/kvm_util_base.h +3 -0
index 66775de26952..0454027d588d 100644
@@ -336,6 +336,9 @@ struct kvm_vm *vm_create_with_vcpus(enum vm_guest_mode mode, uint32_t nr_vcpus,
 				    uint32_t num_percpu_pages, void *guest_code,
 				    uint32_t vcpuids[]);
 
+/* First phase of vm_create_with_vcpus, allows customization before vcpu add */
+struct kvm_vm *vm_create_without_vcpus(enum vm_guest_mode mode, uint64_t pages);
+
 /*
  * Adds a vCPU with reasonable defaults (e.g. a stack)
  *
tools/testing/selftests/kvm/lib/kvm_util.c +35 -13

67 lines not on this page · raw message

Signature · 1 line
2.35.0.rc0.227.g00780c9af4-goog
author
#6 lore raw

PATCH v5 3/3 KVM: selftests: Verify disabling PMU virtualization via KVM_CAP_CONFIG_PMU

In reply to David Dunn

On a VM with PMU disabled via KVM_CAP_PMU_CONFIG, the PMU will not be usable by the guest. On Intel, this causes a #GP. And on AMD, the counters no longer increment.

KVM_CAP_PMU_CONFIG must be invoked on a VM prior to creating VCPUs.

---
 .../kvm/x86_64/pmu_event_filter_test.c        | 35 +++++++++++++++++++
 1 file changed, 35 insertions(+)
tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c +35 -0
index c715adcbd487..7a4b99684d9d 100644
@@ -325,6 +325,39 @@ static void test_not_member_allow_list(struct kvm_vm *vm)
 	TEST_ASSERT(!count, "Disallowed PMU Event is counting");
 }
 
+/*
+ * Verify KVM_CAP_PMU_DISABLE prevents the use of the PMU.
+ *
+ * Note that KVM_CAP_PMU_CAPABILITY must be invoked prior to creating VCPUs.
+ */
+static void test_pmu_config_disable(void (*guest_code)(void))
+{
+	int r;
+	struct kvm_vm *vm;
+	struct kvm_enable_cap cap = { 0 };
+	bool sane;
+
+	r = kvm_check_cap(KVM_CAP_PMU_CAPABILITY);
+	if ((r & KVM_CAP_PMU_DISABLE) == 0)
+		return;
+
+	vm = vm_create_without_vcpus(VM_MODE_DEFAULT, DEFAULT_GUEST_PHY_PAGES);
+
+	cap.cap = KVM_CAP_PMU_CAPABILITY;
+	cap.args[0] = KVM_CAP_PMU_DISABLE;
+	r = vm_enable_cap(vm, &cap);
+	TEST_ASSERT(r == 0, "Failed KVM_CAP_PMU_DISABLE.");
+
+	vm_vcpu_add_default(vm, VCPU_ID, guest_code);
+	vm_init_descriptor_tables(vm);
+	vcpu_init_descriptor_tables(vm, VCPU_ID);
+
+	sane = sanity_check_pmu(vm);
+	TEST_ASSERT(!sane, "Guest should not be able to use disabled PMU.");
+
+	kvm_vm_free(vm);
+}
+
 /*
  * Check for a non-zero PMU version, at least one general-purpose
  * counter per logical processor, an EBX bit vector of length greater
@@ -430,5 +463,7 @@ int main(int argc, char *argv[])
 
 	kvm_vm_free(vm);
 
+	test_pmu_config_disable(guest_code);
+
 	return 0;
 }
Signature · 1 line
2.35.0.rc0.227.g00780c9af4-goog