3 files changed, 24 insertions, 7 deletions

diff --git a/arch/x86/include/asm/perf_event_p4.h b/arch/x86/include/asm/perf_event_p4.h
index e2f6a99f14ab..cc29086e30cd 100644
--- a/arch/x86/include/asm/perf_event_p4.h
+++ b/arch/x86/include/asm/perf_event_p4.h
@@ -22,6 +22,7 @@
 
 #define ARCH_P4_CNTRVAL_BITS    (40)
 #define ARCH_P4_CNTRVAL_MASK    ((1ULL << ARCH_P4_CNTRVAL_BITS) - 1)
+#define ARCH_P4_UNFLAGGED_BIT   ((1ULL) << (ARCH_P4_CNTRVAL_BITS - 1))
 
 #define P4_ESCR_EVENT_MASK      0x7e000000U
 #define P4_ESCR_EVENT_SHIFT     25
diff --git a/arch/x86/kernel/cpu/perf_event_p4.c b/arch/x86/kernel/cpu/perf_event_p4.c
index f7a0993c1e7c..ff751a9f182b 100644
--- a/arch/x86/kernel/cpu/perf_event_p4.c
+++ b/arch/x86/kernel/cpu/perf_event_p4.c
@@ -770,9 +770,14 @@ static inline int p4_pmu_clear_cccr_ovf(struct hw_perf_event *hwc)
                 return 1;
         }
 
-        /* it might be unflagged overflow */
-        rdmsrl(hwc->event_base + hwc->idx, v);
-        if (!(v & ARCH_P4_CNTRVAL_MASK))
+        /*
+         * In some circumstances the overflow might issue an NMI but did
+         * not set P4_CCCR_OVF bit. Because a counter holds a negative value
+         * we simply check for high bit being set, if it's cleared it means
+         * the counter has reached zero value and continued counting before
+         * real NMI signal was received:
+         */
+        if (!(v & ARCH_P4_UNFLAGGED_BIT))
                 return 1;
 
         return 0;
diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 999835b6112b..656222fcf767 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -782,6 +782,10 @@ retry:
         raw_spin_unlock_irq(&ctx->lock);
 }
 
+#define MAX_INTERRUPTS (~0ULL)
+
+static void perf_log_throttle(struct perf_event *event, int enable);
+
 static int
 event_sched_in(struct perf_event *event,
                  struct perf_cpu_context *cpuctx,
@@ -794,6 +798,17 @@ event_sched_in(struct perf_event *event,
 
         event->state = PERF_EVENT_STATE_ACTIVE;
         event->oncpu = smp_processor_id();
+
+        /*
+         * Unthrottle events, since we scheduled we might have missed several
+         * ticks already, also for a heavily scheduling task there is little
+         * guarantee it'll get a tick in a timely manner.
+         */
+        if (unlikely(event->hw.interrupts == MAX_INTERRUPTS)) {
+                perf_log_throttle(event, 1);
+                event->hw.interrupts = 0;
+        }
+
         /*
          * The new state must be visible before we turn it on in the hardware:
          */
@@ -1596,10 +1611,6 @@ void __perf_event_task_sched_in(struct task_struct *task)
         }
 }
 
-#define MAX_INTERRUPTS (~0ULL)
-
-static void perf_log_throttle(struct perf_event *event, int enable);
-
 static u64 perf_calculate_period(struct perf_event *event, u64 nsec, u64 count)
 {
         u64 frequency = event->attr.sample_freq;