async util. monitor

i40 FDIR delay issue

4 files changed, 112 insertions, 17 deletions

diff --git a/scripts/automation/trex_control_plane/stl/trex_stl_lib/trex_stl_async_client.py b/scripts/automation/trex_control_plane/stl/trex_stl_lib/trex_stl_async_client.py
index 745a7fc5..0f73792a 100644
--- a/scripts/automation/trex_control_plane/stl/trex_stl_lib/trex_stl_async_client.py
+++ b/scripts/automation/trex_control_plane/stl/trex_stl_lib/trex_stl_async_client.py
@@ -153,6 +153,8 @@ class CTRexAsyncClient():
         self.last_data_recv_ts = 0
         self.async_barrier     = None
 
+        self.monitor = AsyncUtil()
+
         self.connected = False
  
     # connects the async channel
@@ -223,10 +225,17 @@ class CTRexAsyncClient():
 
         got_data = False
 
+        self.monitor.reset()
+
+
         while self.active:
             try:
 
-                line = self.socket.recv_string()
+                with self.monitor:
+                    line = self.socket.recv_string()
+                
+                self.monitor.on_recv_msg(line)
+
                 self.last_data_recv_ts = time.time()
 
                 # signal once
@@ -237,7 +246,6 @@ class CTRexAsyncClient():
 
             # got a timeout - mark as not alive and retry
             except zmq.Again:
-
                 # signal once
                 if got_data:
                     self.event_handler.on_async_dead()
@@ -341,4 +349,82 @@ class CTRexAsyncClient():
         return RC_OK()
 
 
+# a class to measure util. of async subscriber thread
+class AsyncUtil(object):
+
+    STATE_SLEEP = 1
+    STATE_AWAKE = 2
+
+    def __init__ (self):
+        self.reset()
+
+
+    def reset (self):
+        self.state = self.STATE_AWAKE
+        self.clock = time.time()
+
+        # reset the current interval
+        self.interval = {'ts': time.time(), 'total_sleep': 0, 'total_bits': 0}
+
+        # global counters
+        self.cpu_util = 0
+        self.bps      = 0
+
+
+    def on_recv_msg (self, message):
+        self.interval['total_bits'] += len(message) * 8.0
+
+        self._tick()
+
+
+    def __enter__ (self):
+        assert(self.state == self.STATE_AWAKE)
+        self.state = self.STATE_SLEEP
+
+        self.sleep_start_ts = time.time()
+        
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        assert(self.state == self.STATE_SLEEP)
+        self.state = self.STATE_AWAKE
+
+        # measure total sleep time for interval
+        self.interval['total_sleep'] += time.time() - self.sleep_start_ts
+
+        self._tick()
+                
+    def _tick (self):
+        # how much time did the current interval lasted
+        ts = time.time() - self.interval['ts']
+        if ts < 1:
+            return
+
+        # if tick is in the middle of sleep - add the interval and reset
+        if self.state == self.STATE_SLEEP:
+            self.interval['total_sleep'] += time.time() - self.sleep_start_ts
+            self.sleep_start_ts = time.time()
+
+        # add the interval
+        if self.interval['total_sleep'] > 0:
+            # calculate
+            self.cpu_util = self.cpu_util * 0.75 + (float(ts - self.interval['total_sleep']) / ts) * 0.25
+            self.interval['total_sleep'] = 0
+
+
+        if self.interval['total_bits'] > 0:
+            # calculate
+            self.bps = self.bps * 0.75 + ( self.interval['total_bits'] / ts ) * 0.25
+            self.interval['total_bits'] = 0
+
+        # reset the interval's clock
+        self.interval['ts'] = time.time()
+
+
+    def get_cpu_util (self):
+        self._tick()
+        return (self.cpu_util * 100)
+
+    def get_bps (self):
+        self._tick()
+        return (self.bps)
 
diff --git a/scripts/automation/trex_control_plane/stl/trex_stl_lib/trex_stl_client.py b/scripts/automation/trex_control_plane/stl/trex_stl_lib/trex_stl_client.py
index fa24574f..22895a75 100755
--- a/scripts/automation/trex_control_plane/stl/trex_stl_lib/trex_stl_client.py
+++ b/scripts/automation/trex_control_plane/stl/trex_stl_lib/trex_stl_client.py
@@ -558,7 +558,8 @@ class STLClient(object):
         self.stats_generator = trex_stl_stats.CTRexInfoGenerator(self.global_stats,
                                                                  self.ports,
                                                                  self.flow_stats,
-                                                                 self.latency_stats)
+                                                                 self.latency_stats,
+                                                                 self.async_client.monitor)
 
 
        
diff --git a/scripts/automation/trex_control_plane/stl/trex_stl_lib/trex_stl_stats.py b/scripts/automation/trex_control_plane/stl/trex_stl_lib/trex_stl_stats.py
index 0ebea607..94a45577 100644
--- a/scripts/automation/trex_control_plane/stl/trex_stl_lib/trex_stl_stats.py
+++ b/scripts/automation/trex_control_plane/stl/trex_stl_lib/trex_stl_stats.py
@@ -138,11 +138,12 @@ class CTRexInfoGenerator(object):
     STLClient and the ports.
     """
 
-    def __init__(self, global_stats_ref, ports_dict_ref, rx_stats_ref, latency_stats_ref):
+    def __init__(self, global_stats_ref, ports_dict_ref, rx_stats_ref, latency_stats_ref, async_monitor):
         self._global_stats = global_stats_ref
         self._ports_dict = ports_dict_ref
         self._rx_stats_ref = rx_stats_ref
         self._latency_stats_ref = latency_stats_ref
+        self._async_monitor = async_monitor
 
     def generate_single_statistic(self, port_id_list, statistic_type):
         if statistic_type == GLOBAL_STATS:
@@ -187,24 +188,22 @@ class CTRexInfoGenerator(object):
 
     def _generate_global_stats(self):
         global_stats = self._global_stats
-        #if is_integer(global_stats.max_global_latency):
-        #    max_global_latency_str = ', Max latency: %s' % format_num(global_stats.max_global_latency,
-        #                                                              suffix = 'usec',
-        #                                                              compact = False,
-        #                                                              opts = 'green' if global_stats.max_global_latency < 1000 else 'red')
-        #else:
-        #    max_global_latency_str = ''
+     
         stats_data = OrderedDict([("connection", "{host}, Port {port}".format(host=global_stats.connection_info.get("server"),
                                                                      port=global_stats.connection_info.get("sync_port"))),
                              ("version", "{ver}, UUID: {uuid}".format(ver=global_stats.server_version.get("version", "N/A"),
                                                                       uuid="N/A")),
 
-                             ("cpu_util", "{0}% {1}".format( format_threshold(round_float(global_stats.get("m_cpu_util")), [85, 100], [0, 85]),
+                             ("cpu_util.", "{0}% {1}".format( format_threshold(round_float(global_stats.get("m_cpu_util")), [85, 100], [0, 85]),
                                                               global_stats.get_trend_gui("m_cpu_util", use_raw = True))),
 
-                             ("rx_cpu_util", "{0}% {1}".format( format_threshold(round_float(global_stats.get("m_rx_cpu_util")), [85, 100], [0, 85]),
+                             ("rx_cpu_util.", "{0}% {1}".format( format_threshold(round_float(global_stats.get("m_rx_cpu_util")), [85, 100], [0, 85]),
                                                                 global_stats.get_trend_gui("m_rx_cpu_util", use_raw = True))),
 
+                             ("async_util.", "{0}% / {1}".format( format_threshold(round_float(self._async_monitor.get_cpu_util()), [85, 100], [0, 85]),
+                                                                 format_num(self._async_monitor.get_bps() / 8.0, suffix = "B/sec"))),
+                                                             
+
                              (" ", ""),
 
                              ("total_tx_L2", "{0} {1}".format( global_stats.get("m_tx_bps", format=True, suffix="b/sec"),
diff --git a/src/dpdk22/drivers/net/i40e/i40e_fdir.c b/src/dpdk22/drivers/net/i40e/i40e_fdir.c
index f644ef3d..888b7977 100644
--- a/src/dpdk22/drivers/net/i40e/i40e_fdir.c
+++ b/src/dpdk22/drivers/net/i40e/i40e_fdir.c
@@ -71,8 +71,8 @@
 #define I40E_FDIR_UDP_DEFAULT_LEN           400
 
 /* Wait count and interval for fdir filter programming */
-#define I40E_FDIR_WAIT_COUNT       10
-#define I40E_FDIR_WAIT_INTERVAL_US 1000
+#define I40E_FDIR_WAIT_COUNT       100
+#define I40E_FDIR_WAIT_INTERVAL_US 100
 
 /* Wait count and interval for fdir filter flush */
 #define I40E_FDIR_FLUSH_RETRY       50
@@ -1176,19 +1176,28 @@ i40e_fdir_filter_programming(struct i40e_pf *pf,
 	I40E_PCI_REG_WRITE(txq->qtx_tail, txq->tx_tail);
 
 	for (i = 0; i < I40E_FDIR_WAIT_COUNT; i++) {
-		rte_delay_us(I40E_FDIR_WAIT_INTERVAL_US);
+
+        /* itay: moved this delay after the check to avoid first check */
+        //rte_delay_us(I40E_FDIR_WAIT_INTERVAL_US);
+
 		if ((txdp->cmd_type_offset_bsz &
 				rte_cpu_to_le_64(I40E_TXD_QW1_DTYPE_MASK)) ==
 				rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE))
 			break;
+
+        rte_delay_us(I40E_FDIR_WAIT_INTERVAL_US);
 	}
+
 	if (i >= I40E_FDIR_WAIT_COUNT) {
 		PMD_DRV_LOG(ERR, "Failed to program FDIR filter:"
 			    " time out to get DD on tx queue.");
 		return -ETIMEDOUT;
 	}
+
 	/* totally delay 10 ms to check programming status*/
-	rte_delay_us((I40E_FDIR_WAIT_COUNT - i) * I40E_FDIR_WAIT_INTERVAL_US);
+    /* itay: tests show this is not needed */    
+	//rte_delay_us((I40E_FDIR_WAIT_COUNT - i) * I40E_FDIR_WAIT_INTERVAL_US);
+
 	if (i40e_check_fdir_programming_status(rxq) < 0) {
 		PMD_DRV_LOG(ERR, "Failed to program FDIR filter:"
 			    " programming status reported.");