Add dpdk config parameter: poll-sleep <nn>

Sleep <nn> milliseconds after each dpdk input device poll, useful when
oversubscribing CPUs.

Change-Id: I90ad1f21dae7eeeda56bfe845911118aa46f83ec
Signed-off-by: Dave Barach <dave@barachs.net>

3 files changed, 24 insertions, 21 deletions

diff --git a/vnet/vnet/devices/dpdk/dpdk.h b/vnet/vnet/devices/dpdk/dpdk.h
index 144333fb384..1a861e904f1 100644
--- a/vnet/vnet/devices/dpdk/dpdk.h
+++ b/vnet/vnet/devices/dpdk/dpdk.h
@@ -411,6 +411,9 @@ typedef struct {
   frame_queue_trace_t        *frame_queue_traces;
   frame_queue_nelt_counter_t *frame_queue_histogram;
 
+  /* Sleep for this many MS after each device poll */
+  u32 poll_sleep;
+
   /* convenience */
   vlib_main_t * vlib_main;
   vnet_main_t * vnet_main;
diff --git a/vnet/vnet/devices/dpdk/init.c b/vnet/vnet/devices/dpdk/init.c
index 28659a9d136..3a3c86af49a 100644
--- a/vnet/vnet/devices/dpdk/init.c
+++ b/vnet/vnet/devices/dpdk/init.c
@@ -805,6 +805,7 @@ static clib_error_t *
 dpdk_config (vlib_main_t * vm, unformat_input_t * input)
 {
   clib_error_t * error = 0;
+  dpdk_main_t * dm = &dpdk_main;
   dpdk_config_main_t * conf = &dpdk_config_main;
   vlib_thread_main_t * tm = vlib_get_thread_main();
   vlib_node_runtime_t * rt = vlib_node_get_runtime (vm, dpdk_input_node.index);
@@ -904,7 +905,8 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input)
         conf->use_virtio_vhost = 0;
       else if (unformat (input, "rss %d", &conf->use_rss))
         ;
-
+      else if (unformat (input, "poll-sleep %d", &dm->poll_sleep))
+        ;
 #define _(a)                                    \
       else if (unformat(input, #a))             \
         {                                       \
diff --git a/vnet/vnet/devices/dpdk/node.c b/vnet/vnet/devices/dpdk/node.c
index ca94511b411..0dd3cfcdd77 100644
--- a/vnet/vnet/devices/dpdk/node.c
+++ b/vnet/vnet/devices/dpdk/node.c
@@ -759,24 +759,22 @@ static inline u32 dpdk_device_input ( dpdk_main_t * dm,
   return mb_index;
 }
 
-#if VIRL > 0
-#define VIRL_SPEED_LIMIT()                         \
-  /* Limit the input rate to 1000 vectors / sec */ \
-  {                                                \
-    struct timespec ts, tsrem;                     \
-                                                   \
-    ts.tv_sec = 0;                                 \
-    ts.tv_nsec = 1000*1000; /* 1ms */              \
-                                                   \
-    while (nanosleep(&ts, &tsrem) < 0)             \
-      {                                            \
-        ts = tsrem;                                \
-      }                                            \
-  }
-#else
-#define VIRL_SPEED_LIMIT()
-#endif
+static inline void poll_rate_limit(dpdk_main_t * dm)
+{
+  /* Limit the poll rate by sleeping for N msec between polls */
+  if (PREDICT_FALSE (dm->poll_sleep != 0))
+  {
+    struct timespec ts, tsrem;
+
+    ts.tv_sec = 0;
+    ts.tv_nsec = 1000*1000*dm->poll_sleep; /* 1ms */
 
+    while (nanosleep(&ts, &tsrem) < 0)
+      {
+        ts = tsrem;
+      }
+  }
+}
 
 static uword
 dpdk_input (vlib_main_t * vm,
@@ -799,7 +797,7 @@ dpdk_input (vlib_main_t * vm,
       n_rx_packets += dpdk_device_input (dm, xd, node, cpu_index, 0, 0);
     }
 
-  VIRL_SPEED_LIMIT()
+  poll_rate_limit(dm);
 
   return n_rx_packets;
 }
@@ -824,7 +822,7 @@ dpdk_input_rss (vlib_main_t * vm,
       n_rx_packets += dpdk_device_input (dm, xd, node, cpu_index, dq->queue_id, 0);
     }
 
-  VIRL_SPEED_LIMIT()
+  poll_rate_limit(dm);
 
   return n_rx_packets;
 }
@@ -849,7 +847,7 @@ dpdk_input_efd (vlib_main_t * vm,
       n_rx_packets += dpdk_device_input (dm, xd, node, cpu_index, dq->queue_id, 1);
     }
 
-  VIRL_SPEED_LIMIT()
+  poll_rate_limit(dm);
 
   return n_rx_packets;
 }