[pfcwdorch]: Add alert mode broadcom support (sonic-net#343)

Signed-off-by: Sihui Han <[email protected]>

orchagent/Makefile.am

@@ -6,7 +6,8 @@ swssdir = $(datadir)/swss
 
 dist_swss_DATA = \
 		 pfc_detect_mellanox.lua  \
-		 pfc_restore_mellanox.lua
+		 pfc_detect_broadcom.lua \
+		 pfc_restore.lua
 
 bin_PROGRAMS = orchagent routeresync
 

orchagent/orch.h

@@ -25,6 +25,7 @@ const char comma               = ',';
 const char range_specifier     = '-';
 
 #define MLNX_PLATFORM_SUBSTRING "mellanox"
+#define BRCM_PLATFORM_SUBSTRING "broadcom"
 
 typedef enum
 {

orchagent/orchdaemon.cpp

@@ -124,11 +124,54 @@ bool OrchDaemon::init()
             SAI_QUEUE_STAT_CURR_OCCUPANCY_BYTES,
         };
 
+        static const vector<sai_queue_attr_t> queueAttrIds;
+
         m_orchList.push_back(new PfcWdSwOrch<PfcWdZeroBufferHandler, PfcWdLossyHandler>(
                     m_applDb,
                     pfc_wd_tables,
                     portStatIds,
-                    queueStatIds));
+                    queueStatIds,
+                    queueAttrIds));
+    }
+    else if (platform == BRCM_PLATFORM_SUBSTRING)
+    {
+        static const vector<sai_port_stat_t> portStatIds =
+        {
+            SAI_PORT_STAT_PFC_0_RX_PKTS,
+            SAI_PORT_STAT_PFC_1_RX_PKTS,
+            SAI_PORT_STAT_PFC_2_RX_PKTS,
+            SAI_PORT_STAT_PFC_3_RX_PKTS,
+            SAI_PORT_STAT_PFC_4_RX_PKTS,
+            SAI_PORT_STAT_PFC_5_RX_PKTS,
+            SAI_PORT_STAT_PFC_6_RX_PKTS,
+            SAI_PORT_STAT_PFC_7_RX_PKTS,
+            SAI_PORT_STAT_PFC_0_ON2OFF_RX_PKTS,
+            SAI_PORT_STAT_PFC_1_ON2OFF_RX_PKTS,
+            SAI_PORT_STAT_PFC_2_ON2OFF_RX_PKTS,
+            SAI_PORT_STAT_PFC_3_ON2OFF_RX_PKTS,
+            SAI_PORT_STAT_PFC_4_ON2OFF_RX_PKTS,
+            SAI_PORT_STAT_PFC_5_ON2OFF_RX_PKTS,
+            SAI_PORT_STAT_PFC_6_ON2OFF_RX_PKTS,
+            SAI_PORT_STAT_PFC_7_ON2OFF_RX_PKTS,
+        };
+
+        static const vector<sai_queue_stat_t> queueStatIds =
+        {
+            SAI_QUEUE_STAT_PACKETS,
+            SAI_QUEUE_STAT_CURR_OCCUPANCY_BYTES,
+        };
+
+        static const vector<sai_queue_attr_t> queueAttrIds =
+        {
+            SAI_QUEUE_ATTR_PAUSE_STATUS,
+        };
+
+        m_orchList.push_back(new PfcWdSwOrch<PfcWdActionHandler, PfcWdActionHandler>(
+                    m_applDb,
+                    pfc_wd_tables,
+                    portStatIds,
+                    queueStatIds,
+                    queueAttrIds));
     }
 
     return true;

orchagent/pfc_detect_broadcom.lua

@@ -0,0 +1,91 @@
+-- KEYS - queue IDs
+-- ARGV[1] - counters db index
+-- ARGV[2] - counters table name
+-- ARGV[3] - poll time interval
+-- return queue Ids that satisfy criteria
+
+local counters_db = ARGV[1]
+local counters_table_name = ARGV[2]
+local poll_time = tonumber(ARGV[3])
+
+local rets = {}
+
+redis.call('SELECT', counters_db)
+
+-- Iterate through each queue
+local n = table.getn(KEYS)
+for i = n, 1, -1 do
+    local counter_keys = redis.call('HKEYS', counters_table_name .. ':' .. KEYS[i])
+    local counter_num = 0
+    local old_counter_num = 0
+    local is_deadlock = false
+    local pfc_wd_status = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_STATUS')
+    local pfc_wd_action = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_ACTION')
+    if pfc_wd_status == 'operational' or pfc_wd_action == 'alert' then
+        local detection_time = tonumber(redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_DETECTION_TIME'))
+        local time_left = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_DETECTION_TIME_LEFT')
+        if not time_left  then
+            time_left = detection_time
+        else
+            time_left = tonumber(time_left)
+        end
+
+        local queue_index = redis.call('HGET', 'COUNTERS_QUEUE_INDEX_MAP', KEYS[i])
+        local port_id = redis.call('HGET', 'COUNTERS_QUEUE_PORT_MAP', KEYS[i])
+        local pfc_rx_pkt_key = 'SAI_PORT_STAT_PFC_' .. queue_index .. '_RX_PKTS'
+        local pfc_on2off_key = 'SAI_PORT_STAT_PFC_' .. queue_index .. '_ON2OFF_RX_PKTS'
+
+
+        -- Get all counters
+        local occupancy_bytes = tonumber(redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_CURR_OCCUPANCY_BYTES'))
+        local packets = tonumber(redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS'))
+        local pfc_rx_packets = tonumber(redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key))
+        local pfc_on2off = tonumber(redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_on2off_key))
+        local queue_pause_status = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_ATTR_PAUSE_STATUS')
+
+        local packets_last = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS_last')
+        local pfc_rx_packets_last = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key .. '_last')
+        local pfc_on2off_last = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_on2off_key .. '_last')
+        local queue_pause_status_last = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_ATTR_PAUSE_STATUS_last')
+
+        -- DEBUG CODE START. Uncomment to enable
+        local debug_storm = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'DEBUG_STORM')
+        -- DEBUG CODE END.
+
+        -- If this is not a first run, then we have last values available
+        if packets_last and pfc_rx_packets_last and pfc_on2off_last and queue_pause_status_last then
+            packets_last = tonumber(packets_last)
+            pfc_rx_packets_last = tonumber(pfc_rx_packets_last)
+            pfc_on2off_last = tonumber(pfc_on2off_last)
+
+            -- Check actual condition of queue being in PFC storm
+            if (occupancy_bytes > 0 and packets - packets_last == 0 and pfc_rx_packets - pfc_rx_packets_last > 0) or
+                -- DEBUG CODE START. Uncomment to enable
+                (debug_storm == "enabled") or
+                -- DEBUG CODE END.
+                (occupancy_bytes == 0 and pfc_rx_packets - pfc_rx_packets_last > 0 and pfc_on2off - pfc_on2off_last == 0 and queue_pause_status_last == 'true' and queue_pause_status == 'true') then
+                if time_left <= poll_time then
+                    redis.call('PUBLISH', 'PFC_WD', '["' .. KEYS[i] .. '","storm"]')
+                    is_deadlock = true
+                    time_left = detection_time
+                else
+                    time_left = time_left - poll_time
+                end
+            else
+                if pfc_wd_action == 'alert' and pfc_wd_status ~= 'operational' then
+                    redis.call('PUBLISH', 'PFC_WD', '["' .. KEYS[i] .. '","restore"]')
+                end
+                time_left = detection_time
+            end
+        end
+
+        -- Save values for next run
+        redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_ATTR_PAUSE_STATUS_last', queue_pause_status)
+        redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS_last', packets)
+	redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_DETECTION_TIME_LEFT', time_left)
+        redis.call('HSET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key .. '_last', pfc_rx_packets)
+        redis.call('HSET', counters_table_name .. ':' .. port_id, pfc_on2off_key .. '_last', pfc_on2off)
+    end
+end
+
+return rets

orchagent/pfc_detect_mellanox.lua

@@ -20,7 +20,8 @@ for i = n, 1, -1 do
     local old_counter_num = 0
     local is_deadlock = false
     local pfc_wd_status = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_STATUS')
-    if pfc_wd_status == 'operational' then
+    local pfc_wd_action = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_ACTION')
+    if pfc_wd_status == 'operational' or pfc_wd_action == 'alert' then
         local detection_time = tonumber(redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_DETECTION_TIME'))
         local time_left = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_DETECTION_TIME_LEFT')
         if not time_left  then
@@ -67,6 +68,9 @@ for i = n, 1, -1 do
                     time_left = time_left - poll_time
                 end
             else
+                if pfc_wd_action == 'alert' and pfc_wd_status ~= 'operational' then
+                    redis.call('PUBLISH', 'PFC_WD', '["' .. KEYS[i] .. '","restore"]')
+                end 
                 time_left = detection_time
             end
         end

orchagent/pfc_restore_mellanox.lua → orchagent/pfc_restore.lua

@@ -18,7 +18,8 @@ for i = n, 1, -1 do
     local counter_keys = redis.call('HKEYS', counters_table_name .. ':' .. KEYS[i])
     local pfc_rx_pkt_key = ''
     local pfc_wd_status = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_STATUS')
-    if pfc_wd_status ~= 'operational' then
+    local pfc_wd_action = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_ACTION')
+    if pfc_wd_status ~= 'operational'  and pfc_wd_action ~= 'alert' then
         local restoration_time = tonumber(redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_RESTORATION_TIME'))
         local time_left = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_RESTORATION_TIME_LEFT')
         if time_left == nil then

orchagent/pfcwdorch.cpp

@@ -108,6 +108,27 @@ PfcWdAction PfcWdOrch<DropHandler, ForwardHandler>::deserializeAction(const stri
     return actionMap.at(key);
 }
 
+template <typename DropHandler, typename ForwardHandler>
+string PfcWdOrch<DropHandler, ForwardHandler>::serializeAction(const PfcWdAction& action)
+{
+    SWSS_LOG_ENTER();
+
+    const map<PfcWdAction, string> actionMap =
+    {
+        { PfcWdAction::PFC_WD_ACTION_FORWARD, "forward" },
+        { PfcWdAction::PFC_WD_ACTION_DROP, "drop" },
+        { PfcWdAction::PFC_WD_ACTION_ALERT, "alert" },
+    };
+
+    if (actionMap.find(action) == actionMap.end())
+    {
+        return "unknown";
+    }
+
+    return actionMap.at(action);
+}
+
+
 template <typename DropHandler, typename ForwardHandler>
 void PfcWdOrch<DropHandler, ForwardHandler>::createEntry(const string& key,
         const vector<FieldValueTuple>& data)
@@ -270,6 +291,8 @@ void PfcWdSwOrch<DropHandler, ForwardHandler>::registerInWdDb(const Port& port,
         vector<FieldValueTuple> countersFieldValues;
         countersFieldValues.emplace_back("PFC_WD_DETECTION_TIME", to_string(detectionTime * 1000));
         countersFieldValues.emplace_back("PFC_WD_RESTORATION_TIME", to_string(restorationTime * 1000));
+        countersFieldValues.emplace_back("PFC_WD_ACTION", this->serializeAction(action));
+
         PfcWdOrch<DropHandler, ForwardHandler>::getCountersTable()->set(queueIdStr, countersFieldValues);
 
         // We register our queues in PFC_WD table so that syncd will know that it must poll them
@@ -281,6 +304,12 @@ void PfcWdSwOrch<DropHandler, ForwardHandler>::registerInWdDb(const Port& port,
             queueFieldValues.emplace_back(PFC_WD_QUEUE_COUNTER_ID_LIST, str);
         }
 
+        if (!c_queueAttrIds.empty())
+        {
+            string str = counterIdsToStr(c_queueAttrIds, sai_serialize_queue_attr);
+            queueFieldValues.emplace_back(PFC_WD_QUEUE_ATTR_ID_LIST, str);
+        }
+
         // Create internal entry
         m_entryMap.emplace(queueId, PfcWdQueueEntry(action, port.m_port_id, i));
 
@@ -313,18 +342,19 @@ PfcWdSwOrch<DropHandler, ForwardHandler>::PfcWdSwOrch(
         DBConnector *db,
         vector<string> &tableNames,
         vector<sai_port_stat_t> portStatIds,
-        vector<sai_queue_stat_t> queueStatIds):
+        vector<sai_queue_stat_t> queueStatIds,
+        vector<sai_queue_attr_t> queueAttrIds):
     PfcWdOrch<DropHandler,
     ForwardHandler>(db, tableNames),
     m_pfcWdDb(new DBConnector(PFC_WD_DB, DBConnector::DEFAULT_UNIXSOCKET, 0)),
     m_pfcWdTable(new ProducerStateTable(m_pfcWdDb.get(), PFC_WD_STATE_TABLE)),
     c_portStatIds(portStatIds),
-    c_queueStatIds(queueStatIds)
+    c_queueStatIds(queueStatIds),
+    c_queueAttrIds(queueAttrIds)
 {
     SWSS_LOG_ENTER();
 
     string platform = getenv("platform") ? getenv("platform") : "";
-
     if (platform == "")
     {
         SWSS_LOG_ERROR("Platform environment variable is not defined");
@@ -333,7 +363,7 @@ PfcWdSwOrch<DropHandler, ForwardHandler>::PfcWdSwOrch(
 
     string detectSha, restoreSha;
     string detectPluginName = "pfc_detect_" + platform + ".lua";
-    string restorePluginName = "pfc_restore_" + platform + ".lua";
+    string restorePluginName = "pfc_restore.lua";
 
     try
     {
@@ -428,9 +458,21 @@ void PfcWdSwOrch<DropHandler, ForwardHandler>::handleWdNotification(swss::Notifi
         return;
     }
 
+    SWSS_LOG_NOTICE("Receive notification, %s", event.c_str());
     if (event == "storm")
     {
-        if (entry->second.action == PfcWdAction::PFC_WD_ACTION_DROP)
+        if (entry->second.action == PfcWdAction::PFC_WD_ACTION_ALERT)
+        {
+            if (entry->second.handler == nullptr)
+            {
+                entry->second.handler = make_shared<PfcWdActionHandler>(
+                        entry->second.portId,
+                        entry->first,
+                        entry->second.index,
+                        PfcWdOrch<DropHandler, ForwardHandler>::getCountersTable());
+            }
+        }
+        else if (entry->second.action == PfcWdAction::PFC_WD_ACTION_DROP)
         {
             entry->second.handler = make_shared<DropHandler>(
                     entry->second.portId,
@@ -446,14 +488,6 @@ void PfcWdSwOrch<DropHandler, ForwardHandler>::handleWdNotification(swss::Notifi
                     entry->second.index,
                     PfcWdOrch<DropHandler, ForwardHandler>::getCountersTable());
         }
-        else if (entry->second.action == PfcWdAction::PFC_WD_ACTION_ALERT)
-        {
-            entry->second.handler = make_shared<PfcWdActionHandler>(
-                    entry->second.portId,
-                    entry->first,
-                    entry->second.index,
-                    PfcWdOrch<DropHandler, ForwardHandler>::getCountersTable());
-        }
         else
         {
             throw runtime_error("Unknown PFC WD action");
@@ -465,7 +499,7 @@ void PfcWdSwOrch<DropHandler, ForwardHandler>::handleWdNotification(swss::Notifi
     }
     else
     {
-        SWSS_LOG_ERROR("Received unknown event from plugin");
+        SWSS_LOG_ERROR("Received unknown event from plugin, %s", event.c_str());
     }
 }
 
@@ -551,3 +585,4 @@ void PfcWdSwOrch<DropHandler, ForwardHandler>::endWatchdogThread(void)
 
 // Trick to keep member functions in a separate file
 template class PfcWdSwOrch<PfcWdZeroBufferHandler, PfcWdLossyHandler>;
+template class PfcWdSwOrch<PfcWdActionHandler, PfcWdActionHandler>;

orchagent/pfcwdorch.h

@@ -43,9 +43,10 @@ class PfcWdOrch: public Orch
         return m_countersDb;
     }
 
-private:
     static PfcWdAction deserializeAction(const string& key);
-    void createEntry(const string& key, const vector<FieldValueTuple>& data);
+    static string serializeAction(const PfcWdAction &action); 
+private:
+   void createEntry(const string& key, const vector<FieldValueTuple>& data);
     void deleteEntry(const string& name);
 
     shared_ptr<DBConnector> m_countersDb = nullptr;
@@ -60,7 +61,8 @@ class PfcWdSwOrch: public PfcWdOrch<DropHandler, ForwardHandler>
             DBConnector *db,
             vector<string> &tableNames,
             vector<sai_port_stat_t> portStatIds,
-            vector<sai_queue_stat_t> queueStatIds);
+            vector<sai_queue_stat_t> queueStatIds,
+            vector<sai_queue_attr_t> queueAttrIds);
     virtual ~PfcWdSwOrch(void);
 
     virtual bool startWdOnPort(const Port& port,
@@ -97,6 +99,7 @@ class PfcWdSwOrch: public PfcWdOrch<DropHandler, ForwardHandler>
 
     const vector<sai_port_stat_t> c_portStatIds;
     const vector<sai_queue_stat_t> c_queueStatIds;
+    const vector<sai_queue_attr_t> c_queueAttrIds;
 
     shared_ptr<DBConnector> m_pfcWdDb = nullptr;
     shared_ptr<ProducerStateTable> m_pfcWdTable = nullptr;