perf/x86/core: Add APIs to switch to/from mediated PMI vector (for KVM)

Add APIs (exported only for KVM) to switch PMIs to the dedicated mediated
PMU IRQ vector when loading guest context, and back to perf's standard NMI
when the guest context is put.  I.e. route PMIs to
PERF_GUEST_MEDIATED_PMI_VECTOR when the guest context is active, and to
NMIs while the host context is active.

While running with guest context loaded, ignore all NMIs (in perf).  Any
NMI that arrives while the LVTPC points at the mediated PMU IRQ vector
can't possibly be due to a host perf event.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20251206001720.468579-10-seanjc@google.com

diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c
index 0c38a31d5fc7298b56c95703983f5e13cf66a4b0..3ad5c658e28660d1e98530d322106527820da564 100644 (file)
--- a/arch/x86/events/core.c
+++ b/arch/x86/events/core.c
@@ -56,6 +56,8 @@ DEFINE_PER_CPU(struct cpu_hw_events, cpu_hw_events) = {
        .pmu = &pmu,
 };
 
+static DEFINE_PER_CPU(bool, guest_lvtpc_loaded);
+
 DEFINE_STATIC_KEY_FALSE(rdpmc_never_available_key);
 DEFINE_STATIC_KEY_FALSE(rdpmc_always_available_key);
 DEFINE_STATIC_KEY_FALSE(perf_is_hybrid);
@@ -1760,6 +1762,25 @@ void perf_events_lapic_init(void)
        apic_write(APIC_LVTPC, APIC_DM_NMI);
 }
 
+#ifdef CONFIG_PERF_GUEST_MEDIATED_PMU
+void perf_load_guest_lvtpc(u32 guest_lvtpc)
+{
+       u32 masked = guest_lvtpc & APIC_LVT_MASKED;
+
+       apic_write(APIC_LVTPC,
+                  APIC_DM_FIXED | PERF_GUEST_MEDIATED_PMI_VECTOR | masked);
+       this_cpu_write(guest_lvtpc_loaded, true);
+}
+EXPORT_SYMBOL_FOR_MODULES(perf_load_guest_lvtpc, "kvm");
+
+void perf_put_guest_lvtpc(void)
+{
+       this_cpu_write(guest_lvtpc_loaded, false);
+       apic_write(APIC_LVTPC, APIC_DM_NMI);
+}
+EXPORT_SYMBOL_FOR_MODULES(perf_put_guest_lvtpc, "kvm");
+#endif /* CONFIG_PERF_GUEST_MEDIATED_PMU */
+
 static int
 perf_event_nmi_handler(unsigned int cmd, struct pt_regs *regs)
 {
@@ -1767,6 +1788,17 @@ perf_event_nmi_handler(unsigned int cmd, struct pt_regs *regs)
        u64 finish_clock;
        int ret;
 
+       /*
+        * Ignore all NMIs when the CPU's LVTPC is configured to route PMIs to
+        * PERF_GUEST_MEDIATED_PMI_VECTOR, i.e. when an NMI time can't be due
+        * to a PMI.  Attempting to handle a PMI while the guest's context is
+        * loaded will generate false positives and clobber guest state.  Note,
+        * the LVTPC is switched to/from the dedicated mediated PMI IRQ vector
+        * while host events are quiesced.
+        */
+       if (this_cpu_read(guest_lvtpc_loaded))
+               return NMI_DONE;
+
        /*
         * All PMUs/events that share this PMI handler should make sure to
         * increment active_events for their events.

diff --git a/arch/x86/include/asm/perf_event.h b/arch/x86/include/asm/perf_event.h
index 7276ba70c88a9329330eaa71eb47c60678c4a76b..fb7b261357bfb12c8b5cec6fe1de68e18ab5bbc7 100644 (file)
--- a/arch/x86/include/asm/perf_event.h
+++ b/arch/x86/include/asm/perf_event.h
@@ -759,6 +759,11 @@ static inline void perf_events_lapic_init(void)    { }
 static inline void perf_check_microcode(void) { }
 #endif
 
+#ifdef CONFIG_PERF_GUEST_MEDIATED_PMU
+extern void perf_load_guest_lvtpc(u32 guest_lvtpc);
+extern void perf_put_guest_lvtpc(void);
+#endif
+
 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL)
 extern struct perf_guest_switch_msr *perf_guest_get_msrs(int *nr, void *data);
 extern void x86_perf_get_lbr(struct x86_pmu_lbr *lbr);