Handle situations where another program re-programs the fixed counters

dcdbpusher/sensors/msr/MSRSensorBase.h

@@ -76,7 +76,11 @@ class MSRSensorBase : public SensorBase {
 	void setMetric(uint64_t metric) {
 		_metric = metric;
 	}
-
+	
+	void setFirstReading(bool val) {
+		_firstReading = val;
+	}
+	
 	void printConfig(LOG_LEVEL ll, LOGGER &lg, unsigned leadingSpaces = 16) {
 		std::string leading(leadingSpaces, ' ');
 		LOG_VAR(ll) << leading << "    CPU:               " << _cpu;

dcdbpusher/sensors/msr/MSRSensorGroup.cpp

@@ -131,12 +131,17 @@ void MSRSensorGroup::read() {
 
 	try {
 		for (auto s : _sensors) {
-			auto ret_val = msr_read(s->getMetric(), &reading.value, s->getCpu());
-			if (ret_val != -1) {
-				s->storeReading(reading, 1, !_htAggregation); //1 is no correction...
+			if (checkStatus(s->getMetric(), s->getCpu()) == 0) {
+				auto ret_val = msr_read(s->getMetric(), &reading.value, s->getCpu());
+				if (ret_val != -1) {
+					s->storeReading(reading, 1, !_htAggregation); //1 is no correction...
 #ifdef DEBUG
-				LOG(debug) << _groupName << "::" << s->getName() << " raw reading: \"" << reading.value << "\"";
+					LOG(debug) << _groupName << "::" << s->getName() << " raw reading: \"" << reading.value << "\"";
 #endif
+				}
+			} else {
+				LOG(debug) << _groupName << "::" << s->getName() << " has been disabled, ignoring reading";
+				s->setFirstReading(true);
 			}
 		}
 
@@ -160,8 +165,28 @@ void MSRSensorGroup::read() {
 	} catch (const std::exception &e) {
 		LOG(error) << "Sensorgroup " << _groupName << " could not read value: " << e.what();
 	}
+
+	program_fixed();
 }
 
+int MSRSensorGroup::checkStatus(uint64_t msr_number, unsigned int cpu) {
+	if (!_sensorBins[cpu].isActive()) { // CPU is not active, so it won't be programmed
+		return 0;
+	}
+
+	struct FixedEventControlRegister ctrl_reg;
+	if (msr_read(IA32_CR_FIXED_CTR_CTRL, &ctrl_reg.value, cpu) != -1) {
+		if ((msr_number == INST_RETIRED_ANY_ADDR) && !(ctrl_reg.fields.os0 && ctrl_reg.fields.usr0))
+			return 1;
+		else if ((msr_number == CPU_CLK_UNHALTED_THREAD_ADDR) && !(ctrl_reg.fields.os1 && ctrl_reg.fields.usr1))
+			return 1;
+		else if ((msr_number == CPU_CLK_UNHALTED_REF_ADDR) && !(ctrl_reg.fields.os2 && ctrl_reg.fields.usr2))
+			return 1;
+	}
+
+	return 0;
+}
+	
 int32_t MSRSensorGroup::msr_read(uint64_t msr_number, uint64_t *value, unsigned int cpu) {
 	return pread(_sensorBins[cpu].getFd(), (void *)value, sizeof(uint64_t), msr_number);
 }
@@ -177,7 +202,7 @@ int32_t MSRSensorGroup::msr_write(uint64_t msr_number, uint64_t value, unsigned
  *          because the counters are already in use.
  */
 void MSRSensorGroup::program_fixed() {
-
+	unsigned int freeRunning = 0;
 	for (unsigned int cpu = 0; cpu < _sensorBins.size(); ++cpu) {
 		if (!_sensorBins[cpu].isActive()) { // CPU is not active, so it won't be programmed
 			continue;
@@ -192,7 +217,7 @@ void MSRSensorGroup::program_fixed() {
 		//are they all enabled?
 		if (ctrl_reg.fields.os0 && ctrl_reg.fields.usr0 && ctrl_reg.fields.os1 && ctrl_reg.fields.usr1 && ctrl_reg.fields.os2 && ctrl_reg.fields.usr2) {
 			//yes! Free running counters were set by someone else => we don't need to program them, just read them.
-			LOG(debug) << "CPU" << cpu << " has free running counter, so there will be no fixed counter programming";
+			freeRunning++;
 			continue;
 		}
 		//not all of them (or none) are enabled => we program them again
@@ -225,6 +250,10 @@ void MSRSensorGroup::program_fixed() {
 		//uint64_t value = (1ULL << 32) + (1ULL << 33) + (1ULL << 34);
 		msr_write(IA32_CR_PERF_GLOBAL_CTRL, value, cpu);
 	}
+	
+	if (freeRunning != _sensorBins.size()) {
+		LOG(debug) << "Programmed fixed counters on " << _sensorBins.size()-freeRunning << " CPUs, " << freeRunning << " were already free runnning";
+	}
 }
 
 void MSRSensorGroup::addCpu(unsigned int cpu) {

dcdbpusher/sensors/msr/MSRSensorGroup.h

@@ -99,6 +99,7 @@ class MSRSensorGroup : public SensorGroupTemplate<MSRSensorBase> {
 	};
 
 	void program_fixed();
+	int checkStatus(uint64_t msr_number, unsigned int cpu);
 
 	unsigned int _htAggregation; /**< Value for hyper-threading aggregation. Zero indicates disabled HT agg. */
 	//int _number_metrics_per_cpu;