1 files changed, 133 insertions, 2 deletions

diff --git a/arch/powerpc/perf/core-book3s.c b/arch/powerpc/perf/core-book3s.c
index 73e62e9b179b..b5b42cf0a703 100644
--- a/arch/powerpc/perf/core-book3s.c
+++ b/arch/powerpc/perf/core-book3s.c
@@ -17,6 +17,7 @@
 #include <asm/firmware.h>
 #include <asm/ptrace.h>
 #include <asm/code-patching.h>
+#include <asm/hw_irq.h>
 #include <asm/interrupt.h>
 
 #ifdef CONFIG_PPC64
@@ -775,6 +776,34 @@ static void pmao_restore_workaround(bool ebb)
 	mtspr(SPRN_PMC6, pmcs[5]);
 }
 
+/*
+ * If the perf subsystem wants performance monitor interrupts as soon as
+ * possible (e.g., to sample the instruction address and stack chain),
+ * this should return true. The IRQ masking code can then enable MSR[EE]
+ * in some places (e.g., interrupt handlers) that allows PMI interrupts
+ * through to improve accuracy of profiles, at the cost of some performance.
+ *
+ * The PMU counters can be enabled by other means (e.g., sysfs raw SPR
+ * access), but in that case there is no need for prompt PMI handling.
+ *
+ * This currently returns true if any perf counter is being used. It
+ * could possibly return false if only events are being counted rather than
+ * samples being taken, but for now this is good enough.
+ */
+bool power_pmu_wants_prompt_pmi(void)
+{
+	struct cpu_hw_events *cpuhw;
+
+	/*
+	 * This could simply test local_paca->pmcregs_in_use if that were not
+	 * under ifdef KVM.
+	 */
+	if (!ppmu)
+		return false;
+
+	cpuhw = this_cpu_ptr(&cpu_hw_events);
+	return cpuhw->n_events;
+}
 #endif /* CONFIG_PPC64 */
 
 static void perf_event_interrupt(struct pt_regs *regs);
@@ -857,6 +886,19 @@ static void write_pmc(int idx, unsigned long val)
 	}
 }
 
+static int any_pmc_overflown(struct cpu_hw_events *cpuhw)
+{
+	int i, idx;
+
+	for (i = 0; i < cpuhw->n_events; i++) {
+		idx = cpuhw->event[i]->hw.idx;
+		if ((idx) && ((int)read_pmc(idx) < 0))
+			return idx;
+	}
+
+	return 0;
+}
+
 /* Called from sysrq_handle_showregs() */
 void perf_event_print_debug(void)
 {
@@ -1281,11 +1323,13 @@ static void power_pmu_disable(struct pmu *pmu)
 
 		/*
 		 * Set the 'freeze counters' bit, clear EBE/BHRBA/PMCC/PMAO/FC56
+		 * Also clear PMXE to disable PMI's getting triggered in some
+		 * corner cases during PMU disable.
 		 */
 		val  = mmcr0 = mfspr(SPRN_MMCR0);
 		val |= MMCR0_FC;
 		val &= ~(MMCR0_EBE | MMCR0_BHRBA | MMCR0_PMCC | MMCR0_PMAO |
-			 MMCR0_FC56);
+			 MMCR0_PMXE | MMCR0_FC56);
 		/* Set mmcr0 PMCCEXT for p10 */
 		if (ppmu->flags & PPMU_ARCH_31)
 			val |= MMCR0_PMCCEXT;
@@ -1299,6 +1343,34 @@ static void power_pmu_disable(struct pmu *pmu)
 		mb();
 		isync();
 
+		/*
+		 * Some corner cases could clear the PMU counter overflow
+		 * while a masked PMI is pending. One such case is when
+		 * a PMI happens during interrupt replay and perf counter
+		 * values are cleared by PMU callbacks before replay.
+		 *
+		 * If any PMC corresponding to the active PMU events are
+		 * overflown, disable the interrupt by clearing the paca
+		 * bit for PMI since we are disabling the PMU now.
+		 * Otherwise provide a warning if there is PMI pending, but
+		 * no counter is found overflown.
+		 */
+		if (any_pmc_overflown(cpuhw)) {
+			/*
+			 * Since power_pmu_disable runs under local_irq_save, it
+			 * could happen that code hits a PMC overflow without PMI
+			 * pending in paca. Hence only clear PMI pending if it was
+			 * set.
+			 *
+			 * If a PMI is pending, then MSR[EE] must be disabled (because
+			 * the masked PMI handler disabling EE). So it is safe to
+			 * call clear_pmi_irq_pending().
+			 */
+			if (pmi_irq_pending())
+				clear_pmi_irq_pending();
+		} else
+			WARN_ON(pmi_irq_pending());
+
 		val = mmcra = cpuhw->mmcr.mmcra;
 
 		/*
@@ -1390,6 +1462,15 @@ static void power_pmu_enable(struct pmu *pmu)
 	 * (possibly updated for removal of events).
 	 */
 	if (!cpuhw->n_added) {
+		/*
+		 * If there is any active event with an overflown PMC
+		 * value, set back PACA_IRQ_PMI which would have been
+		 * cleared in power_pmu_disable().
+		 */
+		hard_irq_disable();
+		if (any_pmc_overflown(cpuhw))
+			set_pmi_irq_pending();
+
 		mtspr(SPRN_MMCRA, cpuhw->mmcr.mmcra & ~MMCRA_SAMPLE_ENABLE);
 		mtspr(SPRN_MMCR1, cpuhw->mmcr.mmcr1);
 		if (ppmu->flags & PPMU_ARCH_31)
@@ -2337,6 +2418,14 @@ static void __perf_event_interrupt(struct pt_regs *regs)
 				break;
 			}
 		}
+
+		/*
+		 * Clear PACA_IRQ_PMI in case it was set by
+		 * set_pmi_irq_pending() when PMU was enabled
+		 * after accounting for interrupts.
+		 */
+		clear_pmi_irq_pending();
+
 		if (!active)
 			/* reset non active counters that have overflowed */
 			write_pmc(i + 1, 0);
@@ -2356,6 +2445,13 @@ static void __perf_event_interrupt(struct pt_regs *regs)
 			}
 		}
 	}
+
+	/*
+	 * During system wide profling or while specific CPU is monitored for an
+	 * event, some corner cases could cause PMC to overflow in idle path. This
+	 * will trigger a PMI after waking up from idle. Since counter values are _not_
+	 * saved/restored in idle path, can lead to below "Can't find PMC" message.
+	 */
 	if (unlikely(!found) && !arch_irq_disabled_regs(regs))
 		printk_ratelimited(KERN_WARNING "Can't find PMC that caused IRQ\n");
 
@@ -2392,7 +2488,7 @@ static int power_pmu_prepare_cpu(unsigned int cpu)
 	return 0;
 }
 
-int register_power_pmu(struct power_pmu *pmu)
+int __init register_power_pmu(struct power_pmu *pmu)
 {
 	if (ppmu)
 		return -EBUSY;		/* something's already registered */
@@ -2419,8 +2515,24 @@ int register_power_pmu(struct power_pmu *pmu)
 }
 
 #ifdef CONFIG_PPC64
+static bool pmu_override = false;
+static unsigned long pmu_override_val;
+static void do_pmu_override(void *data)
+{
+	ppc_set_pmu_inuse(1);
+	if (pmu_override_val)
+		mtspr(SPRN_MMCR1, pmu_override_val);
+	mtspr(SPRN_MMCR0, mfspr(SPRN_MMCR0) & ~MMCR0_FC);
+}
+
 static int __init init_ppc64_pmu(void)
 {
+	if (cpu_has_feature(CPU_FTR_HVMODE) && pmu_override) {
+		pr_warn("disabling perf due to pmu_override= command line option.\n");
+		on_each_cpu(do_pmu_override, NULL, 1);
+		return 0;
+	}
+
 	/* run through all the pmu drivers one at a time */
 	if (!init_power5_pmu())
 		return 0;
@@ -2442,4 +2554,23 @@ static int __init init_ppc64_pmu(void)
 		return init_generic_compat_pmu();
 }
 early_initcall(init_ppc64_pmu);
+
+static int __init pmu_setup(char *str)
+{
+	unsigned long val;
+
+	if (!early_cpu_has_feature(CPU_FTR_HVMODE))
+		return 0;
+
+	pmu_override = true;
+
+	if (kstrtoul(str, 0, &val))
+		val = 0;
+
+	pmu_override_val = val;
+
+	return 1;
+}
+__setup("pmu_override=", pmu_setup);
+
 #endif