Skip to contentSkip to conversation

perf_counter: Fix throttling lock-up

tip:perfcounters/core tip-bot for Ingo Molnar 1 participant 1 message Jun 3, 2009 x86

lore
Share & tools
lore
#0 lore raw
Commit-ID:  128f048f0f0d2a477ad2555e7acd2ad15a1b6061
Gitweb:     http://git.kernel.org/tip/128f048f0f0d2a477ad2555e7acd2ad15a1b6061
Author:     Ingo Molnar <[email protected]>
AuthorDate: Wed, 3 Jun 2009 22:19:36 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 3 Jun 2009 23:39:51 +0200

perf_counter: Fix throttling lock-up

Throttling logic is broken and we can lock up with too small hw sampling intervals.

Make the throttling code more robust: disable counters even if we already disabled them.

( Also clean up whitespace damage i noticed while reading
various pieces of code related to throttling. )

Cc: Peter Zijlstra <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Corey Ashford <[email protected]>
Cc: Marcelo Tosatti <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Thomas Gleixner <[email protected]>
LKML-Reference: <new-submission>

---
 arch/x86/kernel/cpu/perf_counter.c |    2 +-
 kernel/perf_counter.c              |   19 ++++++++++++++-----
 2 files changed, 15 insertions(+), 6 deletions(-)
arch/x86/kernel/cpu/perf_counter.c +1 -1
index 12cc05e..8f53f3a 100644
@@ -91,7 +91,7 @@ static u64 intel_pmu_raw_event(u64 event)
 #define CORE_EVNTSEL_INV_MASK		0x00800000ULL
 #define CORE_EVNTSEL_COUNTER_MASK	0xFF000000ULL
 
-#define CORE_EVNTSEL_MASK 		\
+#define CORE_EVNTSEL_MASK		\
 	(CORE_EVNTSEL_EVENT_MASK |	\
 	 CORE_EVNTSEL_UNIT_MASK  |	\
 	 CORE_EVNTSEL_EDGE_MASK  |	\
kernel/perf_counter.c +14 -5
index ab44554..0bb03f1 100644
@@ -2822,11 +2822,20 @@ int perf_counter_overflow(struct perf_counter *counter,
 
 	if (!throttle) {
 		counter->hw.interrupts++;
-	} else if (counter->hw.interrupts != MAX_INTERRUPTS) {
-		counter->hw.interrupts++;
-		if (HZ*counter->hw.interrupts > (u64)sysctl_perf_counter_limit) {
-			counter->hw.interrupts = MAX_INTERRUPTS;
-			perf_log_throttle(counter, 0);
+	} else {
+		if (counter->hw.interrupts != MAX_INTERRUPTS) {
+			counter->hw.interrupts++;
+			if (HZ*counter->hw.interrupts > (u64)sysctl_perf_counter_limit) {
+				counter->hw.interrupts = MAX_INTERRUPTS;
+				perf_log_throttle(counter, 0);
+				ret = 1;
+			}
+		} else {
+			/*
+			 * Keep re-disabling counters even though on the previous
+			 * pass we disabled it - just in case we raced with a
+			 * sched-in and the counter got enabled again:
+			 */
 			ret = 1;
 		}
 	}