[PATCH v7 0/3] KVM: x86: Provide per VM capability for disabling PMU virtualization

[PATCH v7 0/3] KVM: x86: Provide per VM capability for disabling PMU virtualization

[David Dunn]

Allow usermode to disable PMU virtualization on individual x86 VMs.
When disabled, the PMU is not accessible from the guest.

v6: https://lore.kernel.org/all/20220209172945.1495014-1-daviddunn@google.com/

v6 -> v7
 * Incorporating feedback from Sean.
 * Rephrased commit messages (borrowed from Sean).
 * Added locking around kvm->created_vcpus access.
 * Moved KVM_CAP_PMU_VALID_MASK out of user ABI headers.
 * Changed comments on vm_create_without_vcpus.
 * Directly check results in selftest TEST_ASSERT statements.

v5 -> v6
 * resolve minor conflicts that were queued after v5 was reviewed.

v4 -> v5
 * Remove automatic CPUID adjustment when PMU disabled.
 * Update documentation and changelog to reflect above.
 * Update documentation to document arg[0] and return values.

David Dunn (3):
  KVM: x86: Provide per VM capability for disabling PMU virtualization
  KVM: selftests: Carve out helper to create "default" 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                            | 17 +++++++++
 include/uapi/linux/kvm.h                      |  3 ++
 tools/include/uapi/linux/kvm.h                |  3 ++
 .../selftests/kvm/include/kvm_util_base.h     |  3 ++
 tools/testing/selftests/kvm/lib/kvm_util.c    | 35 +++++++++++++++----
 .../kvm/x86_64/pmu_event_filter_test.c        | 33 +++++++++++++++++
 10 files changed, 113 insertions(+), 8 deletions(-)

-- 
2.35.1.265.g69c8d7142f-goog

[PATCH v7 1/3] KVM: x86: Provide per VM capability for disabling PMU virtualization

[David Dunn]

Add a new capability, KVM_CAP_PMU_CAPABILITY, that takes a bitmask of
settings/features to allow userspace to configure PMU virtualization on
a per-VM basis.  For now, support a single flag, KVM_CAP_PMU_DISABLE,
to allow disabling PMU virtualization for a VM even when KVM is configured
with enable_pmu=true a module level.

To keep KVM simple, disallow changing VM's PMU configuration after vCPUs
have been created.

Signed-off-by: David Dunn <daviddunn@google.com>
---
 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              | 17 +++++++++++++++++
 include/uapi/linux/kvm.h        |  3 +++
 tools/include/uapi/linux/kvm.h  |  3 +++
 7 files changed, 48 insertions(+), 2 deletions(-)

diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst
index a4267104db50..df836965b347 100644
--- a/Documentation/virt/kvm/api.rst
+++ b/Documentation/virt/kvm/api.rst
@@ -7561,3 +7561,25 @@ 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.
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 10815b672a26..61e9050a3488 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -1233,6 +1233,7 @@ struct kvm_arch {
 	hpa_t	hv_root_tdp;
 	spinlock_t hv_root_tdp_lock;
 #endif
+	bool enable_pmu;
 };
 
 struct kvm_vm_stat {
diff --git a/arch/x86/kvm/svm/pmu.c b/arch/x86/kvm/svm/pmu.c
index 5aa45f13b16d..d4de52409335 100644
--- a/arch/x86/kvm/svm/pmu.c
+++ b/arch/x86/kvm/svm/pmu.c
@@ -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) {
diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
index 03fab48b149c..4e5b1eeeb77c 100644
--- a/arch/x86/kvm/vmx/pmu_intel.c
+++ b/arch/x86/kvm/vmx/pmu_intel.c
@@ -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;
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index eaa3b5b89c5e..0803a1388bbf 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -110,6 +110,8 @@ static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
 
 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
 
+#define KVM_CAP_PMU_VALID_MASK KVM_CAP_PMU_DISABLE
+
 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
 
@@ -4331,6 +4333,9 @@ 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_VALID_MASK : 0;
+		break;
 	}
 	default:
 		break;
@@ -6005,6 +6010,17 @@ 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 || (cap->args[0] & ~KVM_CAP_PMU_VALID_MASK))
+			break;
+		mutex_lock(&kvm->lock);
+		if (!kvm->created_vcpus) {
+			kvm->arch.enable_pmu = !(cap->args[0] & KVM_CAP_PMU_DISABLE);
+			r = 0;
+		}
+		mutex_unlock(&kvm->lock);
+		break;
 	default:
 		r = -EINVAL;
 		break;
@@ -11591,6 +11607,7 @@ 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);
diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
index 5191b57e1562..3d6046295755 100644
--- a/include/uapi/linux/kvm.h
+++ b/include/uapi/linux/kvm.h
@@ -1134,6 +1134,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_VM_GPA_BITS 207
 #define KVM_CAP_XSAVE2 208
 #define KVM_CAP_SYS_ATTRIBUTES 209
+#define KVM_CAP_PMU_CAPABILITY 210
 
 #ifdef KVM_CAP_IRQ_ROUTING
 
@@ -1970,6 +1971,8 @@ 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)
+
 /**
  * struct kvm_stats_header - Header of per vm/vcpu binary statistics data.
  * @flags: Some extra information for header, always 0 for now.
diff --git a/tools/include/uapi/linux/kvm.h b/tools/include/uapi/linux/kvm.h
index 5191b57e1562..3d6046295755 100644
--- a/tools/include/uapi/linux/kvm.h
+++ b/tools/include/uapi/linux/kvm.h
@@ -1134,6 +1134,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_VM_GPA_BITS 207
 #define KVM_CAP_XSAVE2 208
 #define KVM_CAP_SYS_ATTRIBUTES 209
+#define KVM_CAP_PMU_CAPABILITY 210
 
 #ifdef KVM_CAP_IRQ_ROUTING
 
@@ -1970,6 +1971,8 @@ 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)
+
 /**
  * struct kvm_stats_header - Header of per vm/vcpu binary statistics data.
  * @flags: Some extra information for header, always 0 for now.
-- 
2.35.1.265.g69c8d7142f-goog

[PATCH v7 2/3] KVM: selftests: Carve out helper to create "default" VM without vCPUs

[David Dunn]

Carve out portion of vm_create_default so that selftests can modify
a "default" VM prior to creating vcpus.

Signed-off-by: David Dunn <daviddunn@google.com>
---
 .../selftests/kvm/include/kvm_util_base.h     |  3 ++
 tools/testing/selftests/kvm/lib/kvm_util.c    | 35 +++++++++++++++----
 2 files changed, 32 insertions(+), 6 deletions(-)

diff --git a/tools/testing/selftests/kvm/include/kvm_util_base.h b/tools/testing/selftests/kvm/include/kvm_util_base.h
index 4ed6aa049a91..f987cf7c0d2e 100644
--- a/tools/testing/selftests/kvm/include/kvm_util_base.h
+++ b/tools/testing/selftests/kvm/include/kvm_util_base.h
@@ -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[]);
 
+/* Create a default VM without any vcpus. */
+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)
  *
diff --git a/tools/testing/selftests/kvm/lib/kvm_util.c b/tools/testing/selftests/kvm/lib/kvm_util.c
index d8cf851ab119..5aea7734cfe3 100644
--- a/tools/testing/selftests/kvm/lib/kvm_util.c
+++ b/tools/testing/selftests/kvm/lib/kvm_util.c
@@ -362,6 +362,34 @@ struct kvm_vm *vm_create(enum vm_guest_mode mode, uint64_t phy_pages, int perm)
 	return vm;
 }
 
+/*
+ * Default VM creation without creating VCPUs
+ *
+ * Input Args:
+ *   mode - VM Mode (e.g. VM_MODE_P52V48_4K)
+ *   pages - pages of memory required for VM
+ *
+ * Output Args: None
+ *
+ * Return:
+ *   Pointer to opaque structure that describes the created VM.
+ *
+ * Creates a VM with the mode specified by mode (e.g. VM_MODE_P52V48_4K).
+ */
+struct kvm_vm *vm_create_without_vcpus(enum vm_guest_mode mode, uint64_t pages)
+{
+	struct kvm_vm *vm;
+
+	vm = vm_create(mode, pages, O_RDWR);
+
+	kvm_vm_elf_load(vm, program_invocation_name);
+
+#ifdef __x86_64__
+	vm_create_irqchip(vm);
+#endif
+	return vm;
+}
+
 /*
  * VM Create with customized parameters
  *
@@ -412,13 +440,8 @@ struct kvm_vm *vm_create_with_vcpus(enum vm_guest_mode mode, uint32_t nr_vcpus,
 		    nr_vcpus, kvm_check_cap(KVM_CAP_MAX_VCPUS));
 
 	pages = vm_adjust_num_guest_pages(mode, pages);
-	vm = vm_create(mode, pages, O_RDWR);
-
-	kvm_vm_elf_load(vm, program_invocation_name);
 
-#ifdef __x86_64__
-	vm_create_irqchip(vm);
-#endif
+	vm = vm_create_without_vcpus(mode, pages);
 
 	for (i = 0; i < nr_vcpus; ++i) {
 		uint32_t vcpuid = vcpuids ? vcpuids[i] : i;
-- 
2.35.1.265.g69c8d7142f-goog

[PATCH v7 3/3] KVM: selftests: Verify disabling PMU virtualization via KVM_CAP_CONFIG_PMU

[David Dunn]

On a VM with PMU disabled via KVM_CAP_PMU_CONFIG, the PMU should not be
usable by the guest.

Signed-off-by: David Dunn <daviddunn@google.com>
---
 .../kvm/x86_64/pmu_event_filter_test.c        | 33 +++++++++++++++++++
 1 file changed, 33 insertions(+)

diff --git a/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c b/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c
index c715adcbd487..dc04a266b3ed 100644
--- a/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c
+++ b/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c
@@ -325,6 +325,37 @@ 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 };
+
+	r = kvm_check_cap(KVM_CAP_PMU_CAPABILITY);
+	if (!(r & KVM_CAP_PMU_DISABLE))
+		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;
+	TEST_ASSERT(!vm_enable_cap(vm, &cap), "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);
+
+	TEST_ASSERT(!sanity_check_pmu(vm),
+		    "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 +461,7 @@ int main(int argc, char *argv[])
 
 	kvm_vm_free(vm);
 
+	test_pmu_config_disable(guest_code);
+
 	return 0;
 }
-- 
2.35.1.265.g69c8d7142f-goog

Re: [PATCH v7 2/3] KVM: selftests: Carve out helper to create "default" VM without vCPUs

[Sean Christopherson]

On Tue, Feb 15, 2022, David Dunn wrote:
> Carve out portion of vm_create_default so that selftests can modify
> a "default" VM prior to creating vcpus.
> 
> Signed-off-by: David Dunn <daviddunn@google.com>
> ---
>  .../selftests/kvm/include/kvm_util_base.h     |  3 ++
>  tools/testing/selftests/kvm/lib/kvm_util.c    | 35 +++++++++++++++----
>  2 files changed, 32 insertions(+), 6 deletions(-)
> 
> diff --git a/tools/testing/selftests/kvm/include/kvm_util_base.h b/tools/testing/selftests/kvm/include/kvm_util_base.h
> index 4ed6aa049a91..f987cf7c0d2e 100644
> --- a/tools/testing/selftests/kvm/include/kvm_util_base.h
> +++ b/tools/testing/selftests/kvm/include/kvm_util_base.h
> @@ -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[]);
>  
> +/* Create a default VM without any vcpus. */
> +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)
>   *
> diff --git a/tools/testing/selftests/kvm/lib/kvm_util.c b/tools/testing/selftests/kvm/lib/kvm_util.c
> index d8cf851ab119..5aea7734cfe3 100644
> --- a/tools/testing/selftests/kvm/lib/kvm_util.c
> +++ b/tools/testing/selftests/kvm/lib/kvm_util.c
> @@ -362,6 +362,34 @@ struct kvm_vm *vm_create(enum vm_guest_mode mode, uint64_t phy_pages, int perm)
>  	return vm;
>  }
>  
> +/*
> + * Default VM creation without creating VCPUs
> + *
> + * Input Args:
> + *   mode - VM Mode (e.g. VM_MODE_P52V48_4K)
> + *   pages - pages of memory required for VM
> + *
> + * Output Args: None
> + *
> + * Return:
> + *   Pointer to opaque structure that describes the created VM.
> + *
> + * Creates a VM with the mode specified by mode (e.g. VM_MODE_P52V48_4K).
> + */

I vote to omit this function comment.  Most of the existing comments in kvm_util.c
are a waste of space, and (no offence) this is no different.  And I'm planning on
deleting most of the existing boilerplate comments that don't provide any insight :-)

> +struct kvm_vm *vm_create_without_vcpus(enum vm_guest_mode mode, uint64_t pages)
> +{
> +	struct kvm_vm *vm;
> +
> +	vm = vm_create(mode, pages, O_RDWR);
> +
> +	kvm_vm_elf_load(vm, program_invocation_name);
> +
> +#ifdef __x86_64__
> +	vm_create_irqchip(vm);
> +#endif
> +	return vm;
> +}
> +
>  /*
>   * VM Create with customized parameters
>   *
> @@ -412,13 +440,8 @@ struct kvm_vm *vm_create_with_vcpus(enum vm_guest_mode mode, uint32_t nr_vcpus,
>  		    nr_vcpus, kvm_check_cap(KVM_CAP_MAX_VCPUS));
>  
>  	pages = vm_adjust_num_guest_pages(mode, pages);
> -	vm = vm_create(mode, pages, O_RDWR);
> -
> -	kvm_vm_elf_load(vm, program_invocation_name);
>  
> -#ifdef __x86_64__
> -	vm_create_irqchip(vm);
> -#endif
> +	vm = vm_create_without_vcpus(mode, pages);
>  
>  	for (i = 0; i < nr_vcpus; ++i) {
>  		uint32_t vcpuid = vcpuids ? vcpuids[i] : i;
> -- 
> 2.35.1.265.g69c8d7142f-goog
> 

Re: [PATCH v7 1/3] KVM: x86: Provide per VM capability for disabling PMU virtualization

[Sean Christopherson]

On Tue, Feb 15, 2022, David Dunn wrote:
> Add a new capability, KVM_CAP_PMU_CAPABILITY, that takes a bitmask of
> settings/features to allow userspace to configure PMU virtualization on
> a per-VM basis.  For now, support a single flag, KVM_CAP_PMU_DISABLE,
> to allow disabling PMU virtualization for a VM even when KVM is configured
> with enable_pmu=true a module level.
> 
> To keep KVM simple, disallow changing VM's PMU configuration after vCPUs
> have been created.
> 
> Signed-off-by: David Dunn <daviddunn@google.com>
> ---

A few nits, otherwise

Reviewed-by: Sean Christopherson <seanjc@google.com>

Somewhat off-topic, looking at this got me looking at vmx_get_perf_capabilities().
We really should cache the result of its RDMSR.

And I believe emulation of WRMSR from the guest to MSR_IA32_PERF_CAPABILITIES is
missing a guest_cpuid_has() check.

>  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              | 17 +++++++++++++++++
>  include/uapi/linux/kvm.h        |  3 +++
>  tools/include/uapi/linux/kvm.h  |  3 +++
>  7 files changed, 48 insertions(+), 2 deletions(-)
> 
> diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst
> index a4267104db50..df836965b347 100644
> --- a/Documentation/virt/kvm/api.rst
> +++ b/Documentation/virt/kvm/api.rst
> @@ -7561,3 +7561,25 @@ 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.

Referring to "DISABLE" as a capabilitiy is rather odd, but I don't have a better
suggestion.

> +: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.
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index 10815b672a26..61e9050a3488 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -1233,6 +1233,7 @@ struct kvm_arch {
>  	hpa_t	hv_root_tdp;
>  	spinlock_t hv_root_tdp_lock;
>  #endif
> +	bool enable_pmu;

Doesn''t really matter, but I'd prefer we put this up among the many other bools, e.g.

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 4bb0d3be2055..b57d40419ddd 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -1150,6 +1150,7 @@ struct kvm_arch {
        bool guest_can_read_msr_platform_info;
        bool exception_payload_enabled;
 
+       bool enable_pmu;
        bool bus_lock_detection_enabled;
        /*
         * If exit_on_emulation_error is set, and the in-kernel instruction
@@ -1233,7 +1234,6 @@ struct kvm_arch {
        hpa_t   hv_root_tdp;
        spinlock_t hv_root_tdp_lock;
 #endif
-       bool enable_pmu;
 };
 
 struct kvm_vm_stat {

>  };
>  
>  struct kvm_vm_stat {
> diff --git a/arch/x86/kvm/svm/pmu.c b/arch/x86/kvm/svm/pmu.c
> index 5aa45f13b16d..d4de52409335 100644
> --- a/arch/x86/kvm/svm/pmu.c
> +++ b/arch/x86/kvm/svm/pmu.c
> @@ -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) {
> diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
> index 03fab48b149c..4e5b1eeeb77c 100644
> --- a/arch/x86/kvm/vmx/pmu_intel.c
> +++ b/arch/x86/kvm/vmx/pmu_intel.c
> @@ -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;
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index eaa3b5b89c5e..0803a1388bbf 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -110,6 +110,8 @@ static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
>  
>  #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
>  
> +#define KVM_CAP_PMU_VALID_MASK KVM_CAP_PMU_DISABLE
> +
>  #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
>                                      KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
>  
> @@ -4331,6 +4333,9 @@ 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_VALID_MASK : 0;
> +		break;
>  	}
>  	default:
>  		break;
> @@ -6005,6 +6010,17 @@ 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 || (cap->args[0] & ~KVM_CAP_PMU_VALID_MASK))
> +			break;

I'd prefer another newline here, the other case statements are just bad influences :-)

> +		mutex_lock(&kvm->lock);
> +		if (!kvm->created_vcpus) {
> +			kvm->arch.enable_pmu = !(cap->args[0] & KVM_CAP_PMU_DISABLE);
> +			r = 0;
> +		}
> +		mutex_unlock(&kvm->lock);
> +		break;
>  	default:
>  		r = -EINVAL;
>  		break;