Fixes reconnecting the ZigBee controller after a reset during runtime

Resets may occur by unplugging the stick or calling a softReset() from
the software side.

Additionally resets the stick when 5 subsequent timeouts happen
as there are firmwares of the CC2652 available which tend to
crash occationally.

libnymea-zigbee/backends/ti/interface/zigbeeinterfaceti.cpp

@@ -99,6 +99,7 @@ void ZigbeeInterfaceTi::setAvailable(bool available)
 
 void ZigbeeInterfaceTi::onReconnectTimeout()
 {
+    qCDebug(dcZigbeeInterface()) << "Reconnecting to serial port...";
     if (m_serialPort && !m_serialPort->isOpen()) {
         if (!m_serialPort->open(QSerialPort::ReadWrite)) {
             setAvailable(false);
@@ -169,9 +170,7 @@ void ZigbeeInterfaceTi::onError(const QSerialPort::SerialPortError &error)
 {
     if (error != QSerialPort::NoError && m_serialPort->isOpen()) {
         qCWarning(dcZigbeeInterface()) << "Serial port error:" << error << m_serialPort->errorString();
-        m_reconnectTimer->start();
-        m_serialPort->close();
-        setAvailable(false);
+        reconnectController();
     }
 }
 
@@ -207,7 +206,7 @@ void ZigbeeInterfaceTi::sendPacket(Ti::CommandType type, Ti::SubSystem subSystem
 
 bool ZigbeeInterfaceTi::enable(const QString &serialPort, qint32 baudrate)
 {
-    qCDebug(dcZigbeeInterface()) << "Start UART interface " << serialPort << baudrate;
+    qCDebug(dcZigbeeInterface()) << "Starting UART interface " << serialPort << baudrate;
 
     if (m_serialPort) {
         delete m_serialPort;
@@ -243,13 +242,18 @@ void ZigbeeInterfaceTi::reconnectController()
     if (!m_serialPort)
         return;
 
-    if (m_serialPort->isOpen())
+    if (m_serialPort->isOpen()) {
         m_serialPort->close();
+    }
+
+    QString portName = m_serialPort->portName();
+    int baudrate = m_serialPort->baudRate();
+    setAvailable(false);
 
     delete m_serialPort;
     m_serialPort = nullptr;
-    setAvailable(false);
-    m_reconnectTimer->start();
+
+    enable(portName, baudrate);
 }
 
 void ZigbeeInterfaceTi::disable()

libnymea-zigbee/backends/ti/interface/zigbeeinterfacetireply.cpp

@@ -52,7 +52,7 @@ Ti::StatusCode ZigbeeInterfaceTiReply::statusCode() const
     return m_statusCode;
 }
 
-bool ZigbeeInterfaceTiReply::timendOut() const
+bool ZigbeeInterfaceTiReply::timedOut() const
 {
     return m_timeout;
 }

libnymea-zigbee/backends/ti/interface/zigbeeinterfacetireply.h

@@ -54,7 +54,7 @@ public:
     // Response content
     Ti::StatusCode statusCode() const;
 
-    bool timendOut() const;
+    bool timedOut() const;
     bool aborted() const;
     void abort();
 

libnymea-zigbee/backends/ti/zigbeebridgecontrollerti.cpp

@@ -72,14 +72,22 @@ ZigbeeInterfaceTiReply* ZigbeeBridgeControllerTi::reset()
     stream << static_cast<quint8>(Ti::ResetTypeSoft);
     ZigbeeInterfaceTiReply *resetReply = sendCommand(Ti::SubSystemSys, Ti::SYSCommandResetReq, payload);
     waitFor(resetReply, Ti::SubSystemSys, Ti::SYSCommandResetInd);
+    connect(resetReply, &ZigbeeInterfaceTiReply::finished, this, [=](){
+        m_interface->reconnectController();
+    });
     return resetReply;
 }
 
 ZigbeeInterfaceTiReply *ZigbeeBridgeControllerTi::init()
 {
+    m_registeredEndpointIds.clear();
+
     ZigbeeInterfaceTiReply *initReply = new ZigbeeInterfaceTiReply(this, 15000);
 
-    ZigbeeInterfaceTiReply *resetReply = reset();
+    // Not using public reset() as that will start the init from scratch by reconnecting the controller
+    NEW_PAYLOAD
+    stream << static_cast<quint8>(Ti::ResetTypeSoft);
+    ZigbeeInterfaceTiReply *resetReply = sendCommand(Ti::SubSystemSys, Ti::SYSCommandResetReq, payload);
     connect(resetReply, &ZigbeeInterfaceTiReply::finished, initReply, [=]() {
 
         qCDebug(dcZigbeeController()) << "Skipping CC2530/CC2531 bootloader.";
@@ -486,17 +494,25 @@ void ZigbeeBridgeControllerTi::sendNextRequest()
 
 ZigbeeInterfaceTiReply *ZigbeeBridgeControllerTi::sendCommand(Ti::SubSystem subSystem, quint8 command, const QByteArray &payload, int timeout)
 {
-    // Create the reply
     ZigbeeInterfaceTiReply *reply = new ZigbeeInterfaceTiReply(subSystem, command, this, payload, timeout);
 
-    // Make sure we clean up on timeout
-    connect(reply, &ZigbeeInterfaceTiReply::timeout, this, [reply](){
-        qCWarning(dcZigbeeController()) << "Reply timeout" << reply;
-        // Note: send next reply with the finished signal
-    });
+    connect(reply, &ZigbeeInterfaceTiReply::finished, reply, [=](){
+
+        if (reply->timedOut()) {
+            if (m_controllerState == ControllerStateRunning) {
+                qCWarning(dcZigbeeController()) << "Interface command timed out.";
+                if (++m_timeouts < 5) {
+                    qCInfo(dcZigbeeController()) << "Retrying..." << m_timeouts << "/" << 5;
+                    sendCommand(subSystem, command, payload, timeout);
+                } else {
+                    qCInfo(dcZigbeeController()) << "Resetting ZigBee interface";
+                    m_interface->reconnectController();
+                }
+            }
+        } else {
+            m_timeouts = 0;
+        }
 
-    // Auto delete the object on finished
-    connect(reply, &ZigbeeInterfaceTiReply::finished, reply, [this, reply](){
         if (m_currentReply == reply) {
             m_currentReply = nullptr;
             QMetaObject::invokeMethod(this, "sendNextRequest", Qt::QueuedConnection);
@@ -731,6 +747,9 @@ void ZigbeeBridgeControllerTi::onInterfaceAvailableChanged(bool available)
             ZigbeeInterfaceTiReply *reply = m_replyQueue.dequeue();
             reply->abort();
         }
+
+        m_controllerState = ControllerStateDown;
+        emit controllerStateChanged(m_controllerState);
     }
 
     setAvailable(available);

libnymea-zigbee/backends/ti/zigbeebridgecontrollerti.h

@@ -144,6 +144,7 @@ private:
     QTimer m_permitJoinTimer;
 
     QList<int> m_registeredEndpointIds;
+    int m_timeouts = 0;
 
     void finishRequest(Ti::StatusCode statusCode = Ti::StatusCodeSuccess);
 };

libnymea-zigbee/zcl/zigbeecluster.cpp

@@ -206,6 +206,7 @@ ZigbeeClusterReply *ZigbeeCluster::createClusterReply(const ZigbeeNetworkRequest
     zclReply->m_transactionSequenceNumber = frame.header.transactionSequenceNumber;
     m_pendingReplies.insert(zclReply->transactionSequenceNumber(), zclReply);
     connect(zclReply, &ZigbeeClusterReply::finished, this, [this, zclReply](){
+        qCDebug(dcZigbeeCluster()) << "ZCL request to" << zclReply->request().destinationShortAddress() << "finished with status:" << zclReply->error();
         zclReply->deleteLater();
         m_pendingReplies.remove(zclReply->transactionSequenceNumber());
     });
@@ -367,6 +368,7 @@ bool ZigbeeCluster::verifyNetworkError(ZigbeeClusterReply *zclReply, ZigbeeNetwo
     case ZigbeeNetworkReply::ErrorNoError:
         // The request has been transported successfully to he destination, now
         // wait for the expected indication or check if we already recieved it
+        qCDebug(dcZigbeeCluster()) << "ZCL request sent. Waiting for response data indication...";
         zclReply->m_apsConfirmReceived = true;
         if (!zclReply->m_zclIndicationReceived) {
             zclReply->m_timeoutTimer.start();