cisco-8000 pfc-wd support

orchagent/Makefile.am

@@ -11,7 +11,9 @@ dist_swss_DATA = \
 		 pfc_detect_broadcom.lua \
 		 pfc_detect_barefoot.lua \
 		 pfc_detect_nephos.lua \
+		 pfc_detect_cisco-8000.lua \
 		 pfc_restore.lua \
+		 pfc_restore_cisco-8000.lua \
 		 port_rates.lua \
 		 watermark_queue.lua \
 		 watermark_pg.lua \

orchagent/orch.h

@@ -37,6 +37,7 @@ const char state_db_key_delimiter  = '|';
 #define VS_PLATFORM_SUBSTRING   "vs"
 #define NPS_PLATFORM_SUBSTRING  "nephos"
 #define MRVL_PLATFORM_SUBSTRING "marvell"
+#define CISCO_8000_PLATFORM_SUBSTRING "cisco-8000"
 
 #define CONFIGDB_KEY_SEPARATOR "|"
 #define DEFAULT_KEY_SEPARATOR  ":"

orchagent/orchdaemon.cpp

@@ -520,6 +520,27 @@ bool OrchDaemon::init()
                     queueStatIds,
                     queueAttrIds,
                     PFC_WD_POLL_MSECS));
+    } else if (platform == CISCO_8000_PLATFORM_SUBSTRING)
+    {
+        static const vector<sai_port_stat_t> portStatIds;
+
+        static const vector<sai_queue_stat_t> queueStatIds =
+        {
+            SAI_QUEUE_STAT_PACKETS,
+        };
+
+        static const vector<sai_queue_attr_t> queueAttrIds =
+        {
+            SAI_QUEUE_ATTR_PAUSE_STATUS,
+        };
+
+        m_orchList.push_back(new PfcWdSwOrch<PfcWdSaiDlrInitHandler, PfcWdSaiDlrInitHandler>(
+                    m_configDb,
+                    pfc_wd_tables,
+                    portStatIds,
+                    queueStatIds,
+                    queueAttrIds,
+                    PFC_WD_POLL_MSECS));
     }
 
     m_orchList.push_back(&CounterCheckOrch::getInstance(m_configDb));

orchagent/pfc_detect_cisco-8000.lua

@@ -0,0 +1,76 @@
+-- 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 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')
+    local big_red_switch_mode = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'BIG_RED_SWITCH_MODE')
+    if not big_red_switch_mode and (pfc_wd_status == 'operational' or pfc_wd_action == 'alert') then
+        local detection_time = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_DETECTION_TIME')
+        if detection_time then
+            detection_time = tonumber(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])
+
+            -- Get PFC status
+            local packets = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS')
+            local queue_pause_status = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_ATTR_PAUSE_STATUS')
+
+            if packets and queue_pause_status then
+
+                -- DEBUG CODE START. Uncomment to enable
+                local debug_storm = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'DEBUG_STORM')
+                -- DEBUG CODE END.
+
+                -- Check actual condition of queue being in PFC storm
+                if (queue_pause_status == 'true')
+                    -- DEBUG CODE START. Uncomment to enable
+                    or (debug_storm == "enabled")
+                    -- DEBUG CODE END.
+                    then
+                    if time_left <= poll_time then
+                        redis.call('PUBLISH', 'PFC_WD_ACTION', '["' .. KEYS[i] .. '","storm"]')
+                        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_ACTION', '["' .. KEYS[i] .. '","restore"]')
+                    end
+                    time_left = detection_time
+                end
+
+                -- Save values for next run
+                redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_DETECTION_TIME_LEFT', time_left)
+                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)
+            end
+        end
+    end
+end
+
+return rets

orchagent/pfc_restore_cisco-8000.lua

@@ -0,0 +1,62 @@
+-- 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 pfc_wd_status = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_STATUS')
+    local restoration_time = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_RESTORATION_TIME')
+    local pfc_wd_action = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_ACTION')
+    local big_red_switch_mode = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'BIG_RED_SWITCH_MODE')
+    if not big_red_switch_mode and pfc_wd_status ~= 'operational'  and pfc_wd_action ~= 'alert' and restoration_time and restoration_time ~= '' then
+        restoration_time = tonumber(restoration_time)
+        local time_left = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_RESTORATION_TIME_LEFT')
+        if not time_left then
+            time_left = restoration_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])
+
+        -- DEBUG CODE START. Uncomment to enable
+        local debug_storm = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'DEBUG_STORM')
+        -- DEBUG CODE END.
+
+        -- Check actual condition of queue being restored from PFC storm
+        local queue_pause_status = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_ATTR_PAUSE_STATUS')
+
+        if (queue_pause_status == 'false')
+            -- DEBUG CODE START. Uncomment to enable
+            and (debug_storm ~= "enabled")
+            -- DEBUG CODE END.
+            then
+            if time_left <= 0 then
+                redis.call('PUBLISH', 'PFC_WD_ACTION', '["' .. KEYS[i] .. '","restore"]')
+                time_left = restoration_time
+            else
+                time_left = time_left - poll_time
+            end
+        else
+            time_left = restoration_time
+        end
+
+        -- Save values for next run
+        redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_RESTORATION_TIME_LEFT', time_left)
+    end
+end
+
+return rets

orchagent/pfcactionhandler.cpp

@@ -219,6 +219,82 @@ void PfcWdActionHandler::updateWdCounters(const string& queueIdStr, const PfcWdQ
     m_countersTable->set(queueIdStr, resultFvValues);
 }
 
+PfcWdSaiDlrInitHandler::PfcWdSaiDlrInitHandler(sai_object_id_t port, sai_object_id_t queue,
+                                               uint8_t queueId, shared_ptr<Table> countersTable):
+    PfcWdActionHandler(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;
+    }
+}
+
+PfcWdSaiDlrInitHandler::~PfcWdSaiDlrInitHandler(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;
+    }
+}
+
+bool PfcWdSaiDlrInitHandler::getHwCounters(PfcWdHwStats& counters)
+{
+    SWSS_LOG_ENTER();
+
+    static const vector<sai_stat_id_t> queueStatIds =
+    {
+        SAI_QUEUE_STAT_PACKETS,
+        SAI_QUEUE_STAT_DROPPED_PACKETS,
+    };
+
+    vector<uint64_t> queueStats;
+    queueStats.resize(queueStatIds.size());
+
+    sai_status_t status = sai_queue_api->get_queue_stats(
+            getQueue(),
+            static_cast<uint32_t>(queueStatIds.size()),
+            queueStatIds.data(),
+            queueStats.data());
+
+    if (status != SAI_STATUS_SUCCESS)
+    {
+        SWSS_LOG_ERROR("Failed to fetch queue 0x%" PRIx64 " stats: %d", getQueue(), status);
+        return false;
+    }
+
+    counters.txPkt = queueStats[0];
+    counters.txDropPkt = queueStats[1];
+    counters.rxPkt = 0;
+    counters.rxDropPkt = 0;
+
+    return true;
+}
+
 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

@@ -163,4 +163,15 @@ class PfcWdZeroBufferHandler: public PfcWdLossyHandler
         sai_object_id_t m_originalPgBufferProfile = SAI_NULL_OBJECT_ID;
 };
 
+// PFC queue that implements drop action by draining queue via SAI
+// attribute SAI_QUEUE_ATTR_PFC_DLR_INIT.
+class PfcWdSaiDlrInitHandler: public PfcWdActionHandler
+{
+    public:
+        PfcWdSaiDlrInitHandler(sai_object_id_t port, sai_object_id_t queue,
+                uint8_t queueId, shared_ptr<Table> countersTable);
+        virtual ~PfcWdSaiDlrInitHandler(void);
+        virtual bool getHwCounters(PfcWdHwStats& counters);
+};
+
 #endif

orchagent/pfcwdorch.cpp

@@ -666,7 +666,12 @@ PfcWdSwOrch<DropHandler, ForwardHandler>::PfcWdSwOrch(
 
     string detectSha, restoreSha;
     string detectPluginName = "pfc_detect_" + platform + ".lua";
-    string restorePluginName = "pfc_restore.lua";
+    string restorePluginName;
+    if (platform == CISCO_8000_PLATFORM_SUBSTRING) {
+        restorePluginName = "pfc_restore_" + platform + ".lua";
+    } else {
+        restorePluginName = "pfc_restore.lua";
+    }
 
     try
     {
@@ -1056,3 +1061,4 @@ 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<PfcWdSaiDlrInitHandler, PfcWdSaiDlrInitHandler>;