ASIC internal temperature sensors support (sonic-net#1517)

santhosh-kt · web-flow · commit f6c7422a8a1e · 2021-01-04T10:49:59.000-08:00
**What I did** 1) ASICs have multiple internal thermal sensors. 2) With the help of configuration, a poller is introduced in Orchagent that will periodically retrieve values of these sensors with the help of SAI APIs (opencomputeproject/SAI#880). 3) These retrieved values are being populated to the state DB (In "ASIC_TEMPERATURE_INFO" table). **Why I did it** As part of ASIC Thermal Monitoring HLD.
diff --git a/orchagent/orchdaemon.cpp b/orchagent/orchdaemon.cpp
@@ -74,8 +74,15 @@ bool OrchDaemon::init()
 
     string platform = getenv("platform") ? getenv("platform") : "";
     TableConnector stateDbSwitchTable(m_stateDb, "SWITCH_CAPABILITY");
+    TableConnector app_switch_table(m_applDb, APP_SWITCH_TABLE_NAME);
+    TableConnector conf_asic_sensors(m_configDb, CFG_ASIC_SENSORS_TABLE_NAME);
 
-    gSwitchOrch = new SwitchOrch(m_applDb, APP_SWITCH_TABLE_NAME, stateDbSwitchTable);
+    vector<TableConnector> switch_tables = {
+        conf_asic_sensors,
+        app_switch_table
+    };
+
+    gSwitchOrch = new SwitchOrch(m_applDb, switch_tables, stateDbSwitchTable);
 
     const int portsorch_base_pri = 40;
 
diff --git a/orchagent/switchorch.cpp b/orchagent/switchorch.cpp
@@ -33,17 +33,100 @@ const map<string, sai_packet_action_t> packet_action_map =
     {"trap",    SAI_PACKET_ACTION_TRAP}
 };
 
-SwitchOrch::SwitchOrch(DBConnector *db, string tableName, TableConnector switchTable):
-        Orch(db, tableName),
+SwitchOrch::SwitchOrch(DBConnector *db, vector<TableConnector>& connectors, TableConnector switchTable):
+        Orch(connectors),
         m_switchTable(switchTable.first, switchTable.second),
-        m_db(db)
+        m_db(db),
+        m_stateDb(new DBConnector(STATE_DB, DBConnector::DEFAULT_UNIXSOCKET, 0)),
+        m_asicSensorsTable(new Table(m_stateDb.get(), ASIC_TEMPERATURE_INFO_TABLE_NAME)),
+        m_sensorsPollerTimer (new SelectableTimer((timespec { .tv_sec = DEFAULT_ASIC_SENSORS_POLLER_INTERVAL, .tv_nsec = 0 })))
 {
     m_restartCheckNotificationConsumer = new NotificationConsumer(db, "RESTARTCHECK");
     auto restartCheckNotifier = new Notifier(m_restartCheckNotificationConsumer, this, "RESTARTCHECK");
     Orch::addExecutor(restartCheckNotifier);
+
+    initSensorsTable();
+    auto executorT = new ExecutableTimer(m_sensorsPollerTimer, this, "ASIC_SENSORS_POLL_TIMER");
+    Orch::addExecutor(executorT);
 }
 
-void SwitchOrch::doTask(Consumer &consumer)
+void SwitchOrch::doCfgSensorsTableTask(Consumer &consumer)
+{
+    SWSS_LOG_ENTER();
+
+    auto it = consumer.m_toSync.begin();
+    while (it != consumer.m_toSync.end())
+    {
+        KeyOpFieldsValuesTuple t = it->second;
+        string table_attr = kfvKey(t);
+        string op = kfvOp(t);
+
+        if (op == SET_COMMAND)
+        {
+            FieldValueTuple fvt = kfvFieldsValues(t)[0];
+            SWSS_LOG_NOTICE("ASIC sensors : set %s(%s) to %s", table_attr.c_str(), fvField(fvt).c_str(), fvValue(fvt).c_str());
+
+            if (table_attr == ASIC_SENSORS_POLLER_STATUS)
+            {
+                if (fvField(fvt) == "admin_status")
+                {
+                    if (fvValue(fvt) == "enable" && !m_sensorsPollerEnabled)
+                    {
+                        m_sensorsPollerTimer->start();
+                        m_sensorsPollerEnabled = true;
+                    }
+                    else if (fvValue(fvt) == "disable")
+                    {
+                        m_sensorsPollerEnabled = false;
+                    }
+                    else
+                    {
+                        SWSS_LOG_ERROR("ASIC sensors : unsupported operation for poller state %d",m_sensorsPollerEnabled);
+                    }
+                }
+                else
+                {
+                    SWSS_LOG_ERROR("ASIC sensors : unsupported field in attribute %s", ASIC_SENSORS_POLLER_STATUS);
+                }
+            }
+            else if (table_attr == ASIC_SENSORS_POLLER_INTERVAL)
+            {
+                uint32_t interval=to_uint<uint32_t>(fvValue(fvt));
+
+                if (fvField(fvt) == "interval")
+                {
+                    if (interval != m_sensorsPollerInterval)
+                    {
+                        auto intervT = timespec { .tv_sec = interval , .tv_nsec = 0 };
+                        m_sensorsPollerTimer->setInterval(intervT);
+                        m_sensorsPollerInterval = interval;
+                        m_sensorsPollerIntervalChanged = true;
+                    }
+                    else
+                    {
+                        SWSS_LOG_INFO("ASIC sensors : poller interval unchanged : %d seconds",m_sensorsPollerInterval);
+                    }
+                }
+                else
+                {
+                    SWSS_LOG_ERROR("ASIC sensors : unsupported field in attribute %s", ASIC_SENSORS_POLLER_INTERVAL);
+                }
+            }
+            else
+            {
+                SWSS_LOG_ERROR("ASIC sensors : unsupported attribute %s", table_attr.c_str());
+            }
+        }
+        else
+        {
+            SWSS_LOG_ERROR("ASIC sensors : unsupported operation %s",op.c_str());
+        }
+
+        it = consumer.m_toSync.erase(it);
+    }
+}
+
+void SwitchOrch::doAppSwitchTableTask(Consumer &consumer)
 {
     SWSS_LOG_ENTER();
 
@@ -145,6 +228,26 @@ void SwitchOrch::doTask(Consumer &consumer)
     }
 }
 
+void SwitchOrch::doTask(Consumer &consumer)
+{
+    SWSS_LOG_ENTER();
+    const string & table_name = consumer.getTableName();
+
+    if (table_name == APP_SWITCH_TABLE_NAME)
+    {
+        doAppSwitchTableTask(consumer);
+    }
+    else if (table_name == CFG_ASIC_SENSORS_TABLE_NAME)
+    {
+        doCfgSensorsTableTask(consumer);
+    }
+    else
+    {
+        SWSS_LOG_ERROR("Unknown table : %s", table_name.c_str());
+    }
+
+}
+
 void SwitchOrch::doTask(NotificationConsumer& consumer)
 {
     SWSS_LOG_ENTER();
@@ -209,6 +312,171 @@ bool SwitchOrch::setAgingFDB(uint32_t sec)
     return true;
 }
 
+void SwitchOrch::doTask(SelectableTimer &timer)
+{
+    SWSS_LOG_ENTER();
+
+    if (&timer == m_sensorsPollerTimer)
+    {
+        if (m_sensorsPollerIntervalChanged)
+        {
+            m_sensorsPollerTimer->reset();
+            m_sensorsPollerIntervalChanged = false;
+        }
+
+        if (!m_sensorsPollerEnabled)
+        {
+            m_sensorsPollerTimer->stop();
+            return;
+        }
+
+        sai_attribute_t attr;
+        sai_status_t status;
+        std::vector<FieldValueTuple> values;
+
+        if (m_numTempSensors)
+        {
+            std::vector<int32_t> temp_list(m_numTempSensors);
+
+            memset(&attr, 0, sizeof(attr));
+            attr.id = SAI_SWITCH_ATTR_TEMP_LIST;
+            attr.value.s32list.count = m_numTempSensors;
+            attr.value.s32list.list = temp_list.data();
+
+            status = sai_switch_api->get_switch_attribute(gSwitchId , 1, &attr);
+            if (status == SAI_STATUS_SUCCESS)
+            {
+                for (size_t i = 0; i < attr.value.s32list.count ; i++) {
+                    const std::string &fieldName = "temperature_" + std::to_string(i);
+                    values.emplace_back(fieldName, std::to_string(temp_list[i]));
+                }
+                m_asicSensorsTable->set("",values);
+            }
+            else
+            {
+                SWSS_LOG_ERROR("ASIC sensors : failed to get SAI_SWITCH_ATTR_TEMP_LIST: %d", status);
+            }
+        }
+
+        if (m_sensorsMaxTempSupported)
+        {
+            memset(&attr, 0, sizeof(attr));
+            attr.id = SAI_SWITCH_ATTR_MAX_TEMP;
+
+            status = sai_switch_api->get_switch_attribute(gSwitchId, 1, &attr);
+            if (status == SAI_STATUS_SUCCESS)
+            {
+                const std::string &fieldName = "maximum_temperature";
+                values.emplace_back(fieldName, std::to_string(attr.value.s32));
+                m_asicSensorsTable->set("",values);
+            }
+            else if (status ==  SAI_STATUS_NOT_SUPPORTED || status == SAI_STATUS_NOT_IMPLEMENTED)
+            {
+                m_sensorsMaxTempSupported = false;
+                SWSS_LOG_INFO("ASIC sensors : SAI_SWITCH_ATTR_MAX_TEMP is not supported");
+            }
+            else
+            {
+                m_sensorsMaxTempSupported = false;
+                SWSS_LOG_ERROR("ASIC sensors : failed to get SAI_SWITCH_ATTR_MAX_TEMP: %d", status);
+            }
+        }
+
+        if (m_sensorsAvgTempSupported)
+        {
+            memset(&attr, 0, sizeof(attr));
+            attr.id = SAI_SWITCH_ATTR_AVERAGE_TEMP;
+
+            status = sai_switch_api->get_switch_attribute(gSwitchId, 1, &attr);
+            if (status == SAI_STATUS_SUCCESS)
+            {
+                const std::string &fieldName = "average_temperature";
+                values.emplace_back(fieldName, std::to_string(attr.value.s32));
+                m_asicSensorsTable->set("",values);
+            }
+            else if (status ==  SAI_STATUS_NOT_SUPPORTED || status == SAI_STATUS_NOT_IMPLEMENTED)
+            {
+                m_sensorsAvgTempSupported = false;
+                SWSS_LOG_INFO("ASIC sensors : SAI_SWITCH_ATTR_AVERAGE_TEMP is not supported");
+            }
+            else
+            {
+                m_sensorsAvgTempSupported = false;
+                SWSS_LOG_ERROR("ASIC sensors : failed to get SAI_SWITCH_ATTR_AVERAGE_TEMP: %d", status);
+            }
+        }
+    }
+}
+
+void SwitchOrch::initSensorsTable()
+{
+    SWSS_LOG_ENTER();
+
+    sai_attribute_t attr;
+    sai_status_t status;
+    std::vector<FieldValueTuple> values;
+
+    if (!m_numTempSensorsInitialized)
+    {
+        memset(&attr, 0, sizeof(attr));
+        attr.id = SAI_SWITCH_ATTR_MAX_NUMBER_OF_TEMP_SENSORS;
+
+        status = sai_switch_api->get_switch_attribute(gSwitchId, 1, &attr);
+        if (status == SAI_STATUS_SUCCESS)
+        {
+            m_numTempSensors = attr.value.u8;
+            m_numTempSensorsInitialized = true;
+        }
+        else if (status ==  SAI_STATUS_NOT_SUPPORTED || status == SAI_STATUS_NOT_IMPLEMENTED)
+        {
+            m_numTempSensorsInitialized = true;
+            SWSS_LOG_INFO("ASIC sensors : SAI_SWITCH_ATTR_MAX_NUMBER_OF_TEMP_SENSORS is not supported");
+        }
+        else
+        {
+            SWSS_LOG_ERROR("ASIC sensors : failed to get SAI_SWITCH_ATTR_MAX_NUMBER_OF_TEMP_SENSORS: %d", status);
+        }
+    }
+
+    if (m_numTempSensors)
+    {
+        std::vector<int32_t> temp_list(m_numTempSensors);
+
+        memset(&attr, 0, sizeof(attr));
+        attr.id = SAI_SWITCH_ATTR_TEMP_LIST;
+        attr.value.s32list.count = m_numTempSensors;
+        attr.value.s32list.list = temp_list.data();
+
+        status = sai_switch_api->get_switch_attribute(gSwitchId , 1, &attr);
+        if (status == SAI_STATUS_SUCCESS)
+        {
+            for (size_t i = 0; i < attr.value.s32list.count ; i++) {
+                const std::string &fieldName = "temperature_" + std::to_string(i);
+                values.emplace_back(fieldName, std::to_string(0));
+            }
+            m_asicSensorsTable->set("",values);
+        }
+        else
+        {
+            SWSS_LOG_ERROR("ASIC sensors : failed to get SAI_SWITCH_ATTR_TEMP_LIST: %d", status);
+        }
+    }
+
+    if (m_sensorsMaxTempSupported)
+    {
+        const std::string &fieldName = "maximum_temperature";
+        values.emplace_back(fieldName, std::to_string(0));
+        m_asicSensorsTable->set("",values);
+    }
+
+    if (m_sensorsAvgTempSupported)
+    {
+        const std::string &fieldName = "average_temperature";
+        values.emplace_back(fieldName, std::to_string(0));
+        m_asicSensorsTable->set("",values);
+    }
+}
+
 void SwitchOrch::set_switch_capability(const std::vector<FieldValueTuple>& values)
 {
      m_switchTable.set("switch", values);
diff --git a/orchagent/switchorch.h b/orchagent/switchorch.h
@@ -1,6 +1,11 @@
 #pragma once
 
 #include "orch.h"
+#include "timer.h"
+
+#define DEFAULT_ASIC_SENSORS_POLLER_INTERVAL 60
+#define ASIC_SENSORS_POLLER_STATUS "ASIC_SENSORS_POLLER_STATUS"
+#define ASIC_SENSORS_POLLER_INTERVAL "ASIC_SENSORS_POLLER_INTERVAL"
 
 struct WarmRestartCheck
 {
@@ -12,7 +17,7 @@ struct WarmRestartCheck
 class SwitchOrch : public Orch
 {
 public:
-    SwitchOrch(swss::DBConnector *db, std::string tableName, TableConnector switchTable);
+    SwitchOrch(swss::DBConnector *db, std::vector<TableConnector>& connectors, TableConnector switchTable);
     bool checkRestartReady() { return m_warmRestartCheck.checkRestartReadyState; }
     bool checkRestartNoFreeze() { return m_warmRestartCheck.noFreeze; }
     bool skipPendingTaskCheck() { return m_warmRestartCheck.skipPendingTaskCheck; }
@@ -22,12 +27,28 @@ class SwitchOrch : public Orch
     void set_switch_capability(const std::vector<swss::FieldValueTuple>& values);
 private:
     void doTask(Consumer &consumer);
+    void doTask(swss::SelectableTimer &timer);
+    void doCfgSensorsTableTask(Consumer &consumer);
+    void doAppSwitchTableTask(Consumer &consumer);
+    void initSensorsTable();
 
     swss::NotificationConsumer* m_restartCheckNotificationConsumer;
     void doTask(swss::NotificationConsumer& consumer);
     swss::DBConnector *m_db;
     swss::Table m_switchTable;
 
+    // ASIC temperature sensors
+    std::shared_ptr<swss::DBConnector> m_stateDb = nullptr;
+    std::shared_ptr<swss::Table> m_asicSensorsTable= nullptr;
+    swss::SelectableTimer* m_sensorsPollerTimer = nullptr;
+    bool m_sensorsPollerEnabled = false;
+    uint32_t m_sensorsPollerInterval = DEFAULT_ASIC_SENSORS_POLLER_INTERVAL;
+    bool m_sensorsPollerIntervalChanged = false;
+    uint8_t m_numTempSensors = 0;
+    bool m_numTempSensorsInitialized = false;
+    bool m_sensorsMaxTempSupported = true;
+    bool m_sensorsAvgTempSupported = true;
+
     // Information contained in the request from
     // external program for orchagent pre-shutdown state check
     WarmRestartCheck m_warmRestartCheck = {false, false, false};
diff --git a/tests/mock_tests/aclorch_ut.cpp b/tests/mock_tests/aclorch_ut.cpp
@@ -286,8 +286,16 @@ namespace aclorch_test
             gVirtualRouterId = attr.value.oid;
 
             TableConnector stateDbSwitchTable(m_state_db.get(), "SWITCH_CAPABILITY");
+            TableConnector conf_asic_sensors(m_config_db.get(), CFG_ASIC_SENSORS_TABLE_NAME);
+            TableConnector app_switch_table(m_app_db.get(),  APP_SWITCH_TABLE_NAME);
+
+            vector<TableConnector> switch_tables = {
+                conf_asic_sensors,
+                app_switch_table
+            };
+
             ASSERT_EQ(gSwitchOrch, nullptr);
-            gSwitchOrch = new SwitchOrch(m_app_db.get(), APP_SWITCH_TABLE_NAME, stateDbSwitchTable);
+            gSwitchOrch = new SwitchOrch(m_app_db.get(), switch_tables, stateDbSwitchTable);
 
             // Create dependencies ...
 
