[PATCH 14/23] KVM: x86/pmu, perf/x86: Update effective PMU partition mask

[Zide Chen]

The static system-wide mask x86_pmu.partition_mask defines the global
maximum guest capability, while a guest may have its own mask, which
is a subset of x86_pmu.partition_mask.

The per-guest mask, which on Intel is backed by the PERFMON_MASK VMCS
field, is the effective mask while a vCPU is loaded. Whenever it
changes -- e.g. during kvm_arch_vcpu_{load,put}() -- perf/x86 needs to
be told so it can keep its own per-CPU copy in sync.

Suggested-by: Andi Kleen <ak@xxxxxxxxxxxxxxx>
Signed-off-by: Zide Chen <zide.chen@xxxxxxxxx>
---
 arch/x86/events/core.c            |  8 ++++++++
 arch/x86/include/asm/perf_event.h |  1 +
 arch/x86/kvm/pmu.c                | 13 +++++++++++++
 arch/x86/kvm/pmu.h                |  3 +++
 arch/x86/kvm/x86.c                |  4 ++++
 5 files changed, 29 insertions(+)

diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c
index 282170f7c1e1..f485f927967a 100644
--- a/arch/x86/events/core.c
+++ b/arch/x86/events/core.c
@@ -1883,6 +1883,14 @@ void perf_put_guest_lvtpc(void)
 		apic_write(APIC_LVTPC, APIC_DM_NMI);
 }
 EXPORT_SYMBOL_FOR_KVM(perf_put_guest_lvtpc);
+
+void perf_set_current_partition_mask(u64 mask)
+{
+	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
+
+	cpuc->partition_mask = mask & READ_ONCE(x86_pmu.partition_mask);
+}
+EXPORT_SYMBOL_FOR_KVM(perf_set_current_partition_mask);
 #endif /* CONFIG_PERF_GUEST_MEDIATED_PMU */
 
 static int
diff --git a/arch/x86/include/asm/perf_event.h b/arch/x86/include/asm/perf_event.h
index aaaa34062f8c..5be4f6a93b14 100644
--- a/arch/x86/include/asm/perf_event.h
+++ b/arch/x86/include/asm/perf_event.h
@@ -794,6 +794,7 @@ static inline void perf_check_microcode(void) { }
 extern void perf_pmu_partition_preload(void);
 extern void perf_load_guest_lvtpc(u32 guest_lvtpc);
 extern void perf_put_guest_lvtpc(void);
+extern void perf_set_current_partition_mask(u64 mask);
 #endif
 
 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL)
diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
index f944a15160cb..c447f32c1cb1 100644
--- a/arch/x86/kvm/pmu.c
+++ b/arch/x86/kvm/pmu.c
@@ -46,6 +46,19 @@ EXPORT_SYMBOL_FOR_KVM_INTERNAL(enable_mediated_pmu);
 u64 __read_mostly perfmon_mask;
 EXPORT_SYMBOL_FOR_KVM_INTERNAL(perfmon_mask);
 
+void kvm_pmu_vcpu_load(struct kvm_vcpu *vcpu)
+{
+	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
+
+	perf_set_current_partition_mask(kvm_vcpu_has_perfmon_mask(vcpu) ?
+					pmu->perfmon_mask : 0);
+}
+
+void kvm_pmu_vcpu_put(struct kvm_vcpu *vcpu)
+{
+	perf_set_current_partition_mask(0);
+}
+
 struct kvm_x86_pmu_event_filter {
 	__u32 action;
 	__u32 nevents;
diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h
index 057e3258e473..c5feeb60bcf6 100644
--- a/arch/x86/kvm/pmu.h
+++ b/arch/x86/kvm/pmu.h
@@ -70,6 +70,9 @@ void kvm_pmu_ops_update(const struct kvm_pmu_ops *pmu_ops);
 
 void kvm_handle_guest_mediated_pmi(void);
 
+void kvm_pmu_vcpu_load(struct kvm_vcpu *vcpu);
+void kvm_pmu_vcpu_put(struct kvm_vcpu *vcpu);
+
 static inline bool kvm_pmu_has_perf_global_ctrl(struct kvm_pmu *pmu)
 {
 	/*
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 26a3b7a267b5..07e3fe5b53f8 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -2538,6 +2538,8 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 
 	kvm_request_l1tf_flush_l1d();
 
+	kvm_pmu_vcpu_load(vcpu);
+
 	if (vcpu->scheduled_out && pmu->version && pmu->event_count) {
 		pmu->need_cleanup = true;
 		kvm_make_request(KVM_REQ_PMU, vcpu);
@@ -2686,6 +2688,8 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 
 	kvm_x86_call(vcpu_put)(vcpu);
 	vcpu->arch.last_host_tsc = rdtsc();
+
+	kvm_pmu_vcpu_put(vcpu);
 }
 
 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
-- 
2.55.0