diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 5a3bfa293e8b..defd979003be 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -1484,6 +1484,7 @@ struct kvm_arch {
 
 	bool bus_lock_detection_enabled;
 	bool enable_pmu;
+	bool created_mediated_pmu;
 
 	u32 notify_window;
 	u32 notify_vmexit_flags;
diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
index 487ad19a236e..131e24246b09 100644
--- a/arch/x86/kvm/pmu.c
+++ b/arch/x86/kvm/pmu.c
@@ -135,6 +135,10 @@ void kvm_init_pmu_capability(const struct kvm_pmu_ops *pmu_ops)
 			enable_pmu = false;
 	}
 
+	if (!enable_pmu || !enable_mediated_pmu || !kvm_host_pmu.mediated ||
+	    !pmu_ops->is_mediated_pmu_supported(&kvm_host_pmu))
+		enable_mediated_pmu = false;
+
 	if (!enable_pmu) {
 		memset(&kvm_pmu_cap, 0, sizeof(kvm_pmu_cap));
 		return;
diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h
index 5c3939e91f1d..a5c7c026b919 100644
--- a/arch/x86/kvm/pmu.h
+++ b/arch/x86/kvm/pmu.h
@@ -37,6 +37,8 @@ struct kvm_pmu_ops {
 	void (*deliver_pmi)(struct kvm_vcpu *vcpu);
 	void (*cleanup)(struct kvm_vcpu *vcpu);
 
+	bool (*is_mediated_pmu_supported)(struct x86_pmu_capability *host_pmu);
+
 	const u64 EVENTSEL_EVENT;
 	const int MAX_NR_GP_COUNTERS;
 	const int MIN_NR_GP_COUNTERS;
@@ -58,6 +60,11 @@ static inline bool kvm_pmu_has_perf_global_ctrl(struct kvm_pmu *pmu)
 	return pmu->version > 1;
 }
 
+static inline bool kvm_vcpu_has_mediated_pmu(struct kvm_vcpu *vcpu)
+{
+	return enable_mediated_pmu && vcpu_to_pmu(vcpu)->version;
+}
+
 /*
  * KVM tracks all counters in 64-bit bitmaps, with general purpose counters
  * mapped to bits 31:0 and fixed counters mapped to 63:32, e.g. fixed counter 0
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 1b2827cecf38..fb3a5e861553 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -183,6 +183,10 @@ bool __read_mostly enable_pmu = true;
 EXPORT_SYMBOL_FOR_KVM_INTERNAL(enable_pmu);
 module_param(enable_pmu, bool, 0444);
 
+/* Enable/disabled mediated PMU virtualization. */
+bool __read_mostly enable_mediated_pmu;
+EXPORT_SYMBOL_FOR_KVM_INTERNAL(enable_mediated_pmu);
+
 bool __read_mostly eager_page_split = true;
 module_param(eager_page_split, bool, 0644);
 
@@ -6854,7 +6858,7 @@ int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
 			break;
 
 		mutex_lock(&kvm->lock);
-		if (!kvm->created_vcpus) {
+		if (!kvm->created_vcpus && !kvm->arch.created_mediated_pmu) {
 			kvm->arch.enable_pmu = !(cap->args[0] & KVM_PMU_CAP_DISABLE);
 			r = 0;
 		}
@@ -12641,8 +12645,13 @@ static int sync_regs(struct kvm_vcpu *vcpu)
 	return 0;
 }
 
+#define PERF_MEDIATED_PMU_MSG \
+	"Failed to enable mediated vPMU, try disabling system wide perf events and nmi_watchdog.\n"
+
 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
 {
+	int r;
+
 	if (kvm_check_tsc_unstable() && kvm->created_vcpus)
 		pr_warn_once("SMP vm created on host with unstable TSC; "
 			     "guest TSC will not be reliable\n");
@@ -12653,7 +12662,29 @@ int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
 	if (id >= kvm->arch.max_vcpu_ids)
 		return -EINVAL;
 
-	return kvm_x86_call(vcpu_precreate)(kvm);
+	/*
+	 * Note, any actions done by .vcpu_create() must be idempotent with
+	 * respect to creating multiple vCPUs, and therefore are not undone if
+	 * creating a vCPU fails (including failure during pre-create).
+	 */
+	r = kvm_x86_call(vcpu_precreate)(kvm);
+	if (r)
+		return r;
+
+	if (enable_mediated_pmu && kvm->arch.enable_pmu &&
+	    !kvm->arch.created_mediated_pmu) {
+		if (irqchip_in_kernel(kvm)) {
+			r = perf_create_mediated_pmu();
+			if (r) {
+				pr_warn_ratelimited(PERF_MEDIATED_PMU_MSG);
+				return r;
+			}
+			kvm->arch.created_mediated_pmu = true;
+		} else {
+			kvm->arch.enable_pmu = false;
+		}
+	}
+	return 0;
 }
 
 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
@@ -13319,6 +13350,8 @@ void kvm_arch_destroy_vm(struct kvm *kvm)
 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
 		mutex_unlock(&kvm->slots_lock);
 	}
+	if (kvm->arch.created_mediated_pmu)
+		perf_release_mediated_pmu();
 	kvm_destroy_vcpus(kvm);
 	kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
 #ifdef CONFIG_KVM_IOAPIC
diff --git a/arch/x86/kvm/x86.h b/arch/x86/kvm/x86.h
index fdab0ad49098..6e1fb1680c0a 100644
--- a/arch/x86/kvm/x86.h
+++ b/arch/x86/kvm/x86.h
@@ -470,6 +470,7 @@ extern struct kvm_caps kvm_caps;
 extern struct kvm_host_values kvm_host;
 
 extern bool enable_pmu;
+extern bool enable_mediated_pmu;
 
 /*
  * Get a filtered version of KVM's supported XCR0 that strips out dynamic