PFCWD recovery changes using DLR_INIT

orchagent/orchdaemon.cpp

@@ -597,13 +597,26 @@ bool OrchDaemon::init()
             SAI_QUEUE_ATTR_PAUSE_STATUS,
         };
 
-        m_orchList.push_back(new PfcWdSwOrch<PfcWdAclHandler, PfcWdLossyHandler>(
-                    m_configDb,
-                    pfc_wd_tables,
-                    portStatIds,
-                    queueStatIds,
-                    queueAttrIds,
-                    PFC_WD_POLL_MSECS));
+        if(gSwitchOrch->checkPfcDlrInitEnable())
+        {
+            m_orchList.push_back(new PfcWdSwOrch<PfcWdDlrHandler, PfcWdLossyHandler>(
+                        m_configDb,
+                        pfc_wd_tables,
+                        portStatIds,
+                        queueStatIds,
+                        queueAttrIds,
+                        PFC_WD_POLL_MSECS));
+        }
+        else
+        {
+            m_orchList.push_back(new PfcWdSwOrch<PfcWdAclHandler, PfcWdLossyHandler>(
+                        m_configDb,
+                        pfc_wd_tables,
+                        portStatIds,
+                        queueStatIds,
+                        queueAttrIds,
+                        PFC_WD_POLL_MSECS));
+        }
     } else if (platform == CISCO_8000_PLATFORM_SUBSTRING)
     {
         static const vector<sai_port_stat_t> portStatIds;

orchagent/pfcactionhandler.cpp

@@ -262,6 +262,49 @@ PfcWdSaiDlrInitHandler::~PfcWdSaiDlrInitHandler(void)
     }
 }
 
+PfcWdDlrHandler::PfcWdDlrHandler(sai_object_id_t port, sai_object_id_t queue,
+                                               uint8_t queueId, shared_ptr<Table> countersTable):
+    PfcWdLossyHandler(port, queue, queueId, countersTable)
+{
+    SWSS_LOG_ENTER();
+
+    sai_attribute_t attr;
+    attr.id = SAI_QUEUE_ATTR_PFC_DLR_INIT;
+    attr.value.booldata = true;
+
+    // Set DLR init to true to start PFC deadlock recovery
+    sai_status_t status = sai_queue_api->set_queue_attribute(queue, &attr);
+    if (status != SAI_STATUS_SUCCESS)
+    {
+        SWSS_LOG_ERROR("Failed to set PFC DLR INIT on port 0x%" PRIx64 " queue 0x%" PRIx64
+                       " queueId %d : %d",
+                       port, queue, queueId, status);
+        return;
+    }
+}
+
+PfcWdDlrHandler::~PfcWdDlrHandler(void)
+{
+    SWSS_LOG_ENTER();
+
+    sai_object_id_t port = getPort();
+    sai_object_id_t queue = getQueue();
+    uint8_t queueId = getQueueId();
+
+    sai_attribute_t attr;
+    attr.id = SAI_QUEUE_ATTR_PFC_DLR_INIT;
+    attr.value.booldata = false;
+
+    // Set DLR init to false to stop PFC deadlock recovery
+    sai_status_t status = sai_queue_api->set_queue_attribute(getQueue(), &attr);
+    if (status != SAI_STATUS_SUCCESS)
+    {
+        SWSS_LOG_ERROR("Failed to clear PFC DLR INIT on port 0x%" PRIx64 " queue 0x%" PRIx64
+                       " queueId %d : %d", port, queue, queueId, status);
+        return;
+    }
+}
+
 PfcWdAclHandler::PfcWdAclHandler(sai_object_id_t port, sai_object_id_t queue,
         uint8_t queueId, shared_ptr<Table> countersTable):
     PfcWdLossyHandler(port, queue, queueId, countersTable)

orchagent/pfcactionhandler.h

@@ -115,6 +115,14 @@ class PfcWdAclHandler: public PfcWdLossyHandler
         void updatePfcAclRule(shared_ptr<AclRule> rule, uint8_t queueId, string strTable, vector<sai_object_id_t> port);
 };
 
+class PfcWdDlrHandler: public PfcWdLossyHandler
+{
+    public:
+        PfcWdDlrHandler(sai_object_id_t port, sai_object_id_t queue,
+                uint8_t queueId, shared_ptr<Table> countersTable);
+        virtual ~PfcWdDlrHandler(void);
+};
+
 // PFC queue that implements drop action by draining queue with buffer of zero size
 class PfcWdZeroBufferHandler: public PfcWdLossyHandler
 {

orchagent/pfcwdorch.cpp

@@ -27,9 +27,12 @@
 #define PFC_WD_TC_MAX 8
 #define COUNTER_CHECK_POLL_TIMEOUT_SEC  1
 
+extern sai_object_id_t gSwitchId;
+extern sai_switch_api_t* sai_switch_api;
 extern sai_port_api_t *sai_port_api;
 extern sai_queue_api_t *sai_queue_api;
 
+extern SwitchOrch *gSwitchOrch;
 extern PortsOrch *gPortsOrch;
 
 template <typename DropHandler, typename ForwardHandler>
@@ -229,6 +232,36 @@ task_process_status PfcWdOrch<DropHandler, ForwardHandler>::createEntry(const st
                     SWSS_LOG_ERROR("Unsupported action %s for platform %s", value.c_str(), m_platform.c_str());
                     return task_process_status::task_invalid_entry;
                 }
+                if(m_platform == BRCM_PLATFORM_SUBSTRING)
+                {
+                    if(gSwitchOrch->checkPfcDlrInitEnable()) 
+                    {
+                        if(getPfcDlrPacketAction() == PfcWdAction::PFC_WD_ACTION_UNKNOWN) 
+                        {
+                            sai_attribute_t attr;
+                            attr.id = SAI_SWITCH_ATTR_PFC_DLR_PACKET_ACTION;
+                            attr.value.u32 = (sai_uint32_t)action;
+
+                            sai_status_t status = sai_switch_api->set_switch_attribute(gSwitchId, &attr);
+                            if(status != SAI_STATUS_SUCCESS)
+                            {
+                                SWSS_LOG_ERROR("Failed to set switch level PFC DLR packet action rv : %d", status);
+                                return task_process_status::task_invalid_entry;
+                            }
+                            setPfcDlrPacketAction(action);
+                        }
+                        else
+                        {
+                            if(getPfcDlrPacketAction() != action) 
+                            {
+                                string DlrPacketAction = serializeAction(getPfcDlrPacketAction());
+                                SWSS_LOG_ERROR("Invalid PFC Watchdog action %s as switch level action %s is set", 
+                                               value.c_str(), DlrPacketAction.c_str());
+                                return task_process_status::task_invalid_entry;
+                            }
+                        }
+                    }
+                }
             }
             else
             {
@@ -1064,4 +1097,5 @@ bool PfcWdSwOrch<DropHandler, ForwardHandler>::bake()
 // Trick to keep member functions in a separate file
 template class PfcWdSwOrch<PfcWdZeroBufferHandler, PfcWdLossyHandler>;
 template class PfcWdSwOrch<PfcWdAclHandler, PfcWdLossyHandler>;
+template class PfcWdSwOrch<PfcWdDlrHandler, PfcWdLossyHandler>;
 template class PfcWdSwOrch<PfcWdSaiDlrInitHandler, PfcWdActionHandler>;

orchagent/pfcwdorch.h

@@ -49,6 +49,8 @@ class PfcWdOrch: public Orch
 
     virtual task_process_status createEntry(const string& key, const vector<FieldValueTuple>& data);
     task_process_status deleteEntry(const string& name);
+    PfcWdAction getPfcDlrPacketAction() { return PfcDlrPacketAction; }
+    void setPfcDlrPacketAction(PfcWdAction action) { PfcDlrPacketAction = action; }
 
 protected:
     virtual bool startWdActionOnQueue(const string &event, sai_object_id_t queueId) = 0;
@@ -58,6 +60,7 @@ class PfcWdOrch: public Orch
 
     shared_ptr<DBConnector> m_countersDb = nullptr;
     shared_ptr<Table> m_countersTable = nullptr;
+    PfcWdAction PfcDlrPacketAction = PfcWdAction::PFC_WD_ACTION_UNKNOWN;
 };
 
 template <typename DropHandler, typename ForwardHandler>

orchagent/qosorch.cpp

@@ -1670,7 +1670,7 @@ bool QosOrch::applyDscpToTcMapToSwitch(sai_attr_id_t attr_id, sai_object_id_t ma
     SWSS_LOG_ENTER();
 
     /* Query DSCP_TO_TC QoS map at switch capability */
-    bool rv = gSwitchOrch->querySwitchDscpToTcCapability(SAI_OBJECT_TYPE_SWITCH, SAI_SWITCH_ATTR_QOS_DSCP_TO_TC_MAP);
+    bool rv = gSwitchOrch->querySwitchCapability(SAI_OBJECT_TYPE_SWITCH, SAI_SWITCH_ATTR_QOS_DSCP_TO_TC_MAP);
     if (rv == false)
     {
         SWSS_LOG_ERROR("Switch level DSCP to TC QoS map configuration is not supported");

orchagent/switchorch.cpp

@@ -48,6 +48,27 @@ const map<string, sai_packet_action_t> packet_action_map =
 
 const std::set<std::string> switch_non_sai_attribute_set = {"ordered_ecmp"};
 
+void SwitchOrch::set_switch_pfc_dlr_init_capability()
+{
+    vector<FieldValueTuple> fvVector;
+
+    /* Query PFC DLR INIT capability */
+    bool rv = querySwitchCapability(SAI_OBJECT_TYPE_QUEUE, SAI_QUEUE_ATTR_PFC_DLR_INIT);
+    if (rv == false)
+    {
+        SWSS_LOG_INFO("Queue level PFC DLR INIT configuration is not supported");
+        m_PfcDlrInitEnable = false;
+        fvVector.emplace_back(SWITCH_CAPABILITY_TABLE_PFC_DLR_INIT_CAPABLE, "false");
+    }
+    else 
+    {
+        SWSS_LOG_INFO("Queue level PFC DLR INIT configuration is supported");
+        m_PfcDlrInitEnable = true;
+        fvVector.emplace_back(SWITCH_CAPABILITY_TABLE_PFC_DLR_INIT_CAPABLE, "true");
+    }
+    set_switch_capability(fvVector);
+}
+
 SwitchOrch::SwitchOrch(DBConnector *db, vector<TableConnector>& connectors, TableConnector switchTable):
         Orch(connectors),
         m_switchTable(switchTable.first, switchTable.second),
@@ -60,6 +81,7 @@ SwitchOrch::SwitchOrch(DBConnector *db, vector<TableConnector>& connectors, Tabl
     auto restartCheckNotifier = new Notifier(m_restartCheckNotificationConsumer, this, "RESTARTCHECK");
     Orch::addExecutor(restartCheckNotifier);
 
+    set_switch_pfc_dlr_init_capability();
     initSensorsTable();
     querySwitchTpidCapability();
     auto executorT = new ExecutableTimer(m_sensorsPollerTimer, this, "ASIC_SENSORS_POLL_TIMER");
@@ -762,7 +784,7 @@ void SwitchOrch::querySwitchTpidCapability()
     }
 }
 
-bool SwitchOrch::querySwitchDscpToTcCapability(sai_object_type_t sai_object, sai_attr_id_t attr_id)
+bool SwitchOrch::querySwitchCapability(sai_object_type_t sai_object, sai_attr_id_t attr_id)
 {
     SWSS_LOG_ENTER();
 

orchagent/switchorch.h

@@ -11,6 +11,7 @@
 #define SWITCH_CAPABILITY_TABLE_PORT_TPID_CAPABLE                      "PORT_TPID_CAPABLE"
 #define SWITCH_CAPABILITY_TABLE_LAG_TPID_CAPABLE                       "LAG_TPID_CAPABLE"
 #define SWITCH_CAPABILITY_TABLE_ORDERED_ECMP_CAPABLE                   "ORDERED_ECMP_CAPABLE"
+#define SWITCH_CAPABILITY_TABLE_PFC_DLR_INIT_CAPABLE                   "PFC_DLR_INIT_CAPABLE"
 
 struct WarmRestartCheck
 {
@@ -30,7 +31,9 @@ class SwitchOrch : public Orch
     void restartCheckReply(const std::string &op, const std::string &data, std::vector<swss::FieldValueTuple> &values);
     bool setAgingFDB(uint32_t sec);
     void set_switch_capability(const std::vector<swss::FieldValueTuple>& values);
-    bool querySwitchDscpToTcCapability(sai_object_type_t sai_object, sai_attr_id_t attr_id);
+    bool querySwitchCapability(sai_object_type_t sai_object, sai_attr_id_t attr_id);
+    bool checkPfcDlrInitEnable() { return m_PfcDlrInitEnable; }
+    void set_switch_pfc_dlr_init_capability();
 
     // Return reference to ACL group created for each stage and the bind point is
     // the switch
@@ -80,6 +83,7 @@ class SwitchOrch : public Orch
     bool m_sensorsAvgTempSupported = true;
     bool m_vxlanSportUserModeEnabled = false;
     bool m_orderedEcmpEnable = false;
+    bool m_PfcDlrInitEnable = false;
 
     // Information contained in the request from
     // external program for orchagent pre-shutdown state check