Cleanup of handoff code

- removed handoff-dispatch node
- removed some unused buffer metadata fields
- enqueue to thread logic moved to inline function

Change-Id: I7361e1d88f8cce74cd4fcec90d172eade1855cbd
Signed-off-by: Damjan Marion <damarion@cisco.com>

5 files changed, 24 insertions, 341 deletions

diff --git a/src/vnet/buffer.h b/src/vnet/buffer.h
index 9d45c7ee354..b4ce9c79c87 100644
--- a/src/vnet/buffer.h
+++ b/src/vnet/buffer.h
@@ -51,7 +51,6 @@
   _( 3, VLAN_2_DEEP, "vlan-2-deep")			\
   _( 4, VLAN_1_DEEP, "vlan-1-deep")			\
   _( 5, SPAN_CLONE, "span-clone")			\
-  _( 6, HANDOFF_NEXT_VALID, "handoff-next-valid")	\
   _( 7, LOCALLY_ORIGINATED, "local")			\
   _( 8, IS_IP4, "ip4")					\
   _( 9, IS_IP6, "ip6")					\
@@ -88,7 +87,6 @@ _(ip)                                           \
 _(l2)                                           \
 _(l2t)                                          \
 _(l2_classify)                                  \
-_(handoff)                                      \
 _(policer)                                      \
 _(ipsec)					\
 _(map)						\
@@ -239,12 +237,6 @@ typedef struct
       u64 hash;
     } l2_classify;
 
-    /* IO - worker thread handoff */
-    struct
-    {
-      u32 next_index;
-    } handoff;
-
     /* vnet policer */
     struct
     {
@@ -304,13 +296,6 @@ typedef struct
       u16 overlay_afi;
     } lisp;
 
-    /* Driver rx feature */
-    struct
-    {
-      u32 saved_next_index;		/**< saved by drivers for short-cut */
-      u16 buffer_advance;
-    } device_input_feat;
-
     /* TCP */
     struct
     {
diff --git a/src/vnet/feature/feature.h b/src/vnet/feature/feature.h
index 81224ebf0e2..66ee97c87d6 100644
--- a/src/vnet/feature/feature.h
+++ b/src/vnet/feature/feature.h
@@ -265,9 +265,7 @@ vnet_feature_start_device_input_x1 (u32 sw_if_index, u32 * next0,
        */
       u16 adv;
 
-      vnet_buffer (b0)->device_input_feat.saved_next_index = *next0;
       adv = device_input_next_node_advance[*next0];
-      vnet_buffer (b0)->device_input_feat.buffer_advance = adv;
       vlib_buffer_advance (b0, -adv);
 
       vnet_buffer (b0)->feature_arc_index = feature_arc_index;
@@ -299,14 +297,10 @@ vnet_feature_start_device_input_x2 (u32 sw_if_index,
        */
       u16 adv;
 
-      vnet_buffer (b0)->device_input_feat.saved_next_index = *next0;
       adv = device_input_next_node_advance[*next0];
-      vnet_buffer (b0)->device_input_feat.buffer_advance = adv;
       vlib_buffer_advance (b0, -adv);
 
-      vnet_buffer (b1)->device_input_feat.saved_next_index = *next1;
       adv = device_input_next_node_advance[*next1];
-      vnet_buffer (b1)->device_input_feat.buffer_advance = adv;
       vlib_buffer_advance (b1, -adv);
 
       vnet_buffer (b0)->feature_arc_index = feature_arc_index;
@@ -346,24 +340,16 @@ vnet_feature_start_device_input_x4 (u32 sw_if_index,
        */
       u16 adv;
 
-      vnet_buffer (b0)->device_input_feat.saved_next_index = *next0;
       adv = device_input_next_node_advance[*next0];
-      vnet_buffer (b0)->device_input_feat.buffer_advance = adv;
       vlib_buffer_advance (b0, -adv);
 
-      vnet_buffer (b1)->device_input_feat.saved_next_index = *next1;
       adv = device_input_next_node_advance[*next1];
-      vnet_buffer (b1)->device_input_feat.buffer_advance = adv;
       vlib_buffer_advance (b1, -adv);
 
-      vnet_buffer (b2)->device_input_feat.saved_next_index = *next2;
       adv = device_input_next_node_advance[*next2];
-      vnet_buffer (b2)->device_input_feat.buffer_advance = adv;
       vlib_buffer_advance (b2, -adv);
 
-      vnet_buffer (b3)->device_input_feat.saved_next_index = *next3;
       adv = device_input_next_node_advance[*next3];
-      vnet_buffer (b3)->device_input_feat.buffer_advance = adv;
       vlib_buffer_advance (b3, -adv);
 
       vnet_buffer (b0)->feature_arc_index = feature_arc_index;
diff --git a/src/vnet/handoff.c b/src/vnet/handoff.c
index 81cb9f55fba..11b877e500d 100644
--- a/src/vnet/handoff.c
+++ b/src/vnet/handoff.c
@@ -37,15 +37,10 @@ typedef struct
   /* Worker handoff index */
   u32 frame_queue_index;
 
-  /* convenience variables */
-  vlib_main_t *vlib_main;
-  vnet_main_t *vnet_main;
-
     u64 (*hash_fn) (ethernet_header_t *);
 } handoff_main_t;
 
 handoff_main_t handoff_main;
-vlib_node_registration_t handoff_dispatch_node;
 
 typedef struct
 {
@@ -75,145 +70,68 @@ worker_handoff_node_fn (vlib_main_t * vm,
 			vlib_node_runtime_t * node, vlib_frame_t * frame)
 {
   handoff_main_t *hm = &handoff_main;
-  vlib_thread_main_t *tm = vlib_get_thread_main ();
+  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
   u32 n_left_from, *from;
-  static __thread vlib_frame_queue_elt_t **handoff_queue_elt_by_worker_index;
-  static __thread vlib_frame_queue_t **congested_handoff_queue_by_worker_index
-    = 0;
-  vlib_frame_queue_elt_t *hf = 0;
-  int i;
-  u32 n_left_to_next_worker = 0, *to_next_worker = 0;
-  u32 next_worker_index = 0;
-  u32 current_worker_index = ~0;
-
-  if (PREDICT_FALSE (handoff_queue_elt_by_worker_index == 0))
-    {
-      vec_validate (handoff_queue_elt_by_worker_index, tm->n_vlib_mains - 1);
-
-      vec_validate_init_empty (congested_handoff_queue_by_worker_index,
-			       hm->first_worker_index + hm->num_workers - 1,
-			       (vlib_frame_queue_t *) (~0));
-    }
+  u16 thread_indices[VLIB_FRAME_SIZE], *ti;
 
   from = vlib_frame_vector_args (frame);
   n_left_from = frame->n_vectors;
+  vlib_get_buffers (vm, from, bufs, n_left_from);
+
+  b = bufs;
+  ti = thread_indices;
 
   while (n_left_from > 0)
     {
-      u32 bi0;
-      vlib_buffer_t *b0;
       u32 sw_if_index0;
       u32 hash;
       u64 hash_key;
       per_inteface_handoff_data_t *ihd0;
       u32 index0;
 
-      bi0 = from[0];
-      from += 1;
-      n_left_from -= 1;
 
-      b0 = vlib_get_buffer (vm, bi0);
-      sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+      sw_if_index0 = vnet_buffer (b[0])->sw_if_index[VLIB_RX];
       ASSERT (hm->if_data);
       ihd0 = vec_elt_at_index (hm->if_data, sw_if_index0);
 
-      next_worker_index = hm->first_worker_index;
-
       /*
        * Force unknown traffic onto worker 0,
        * and into ethernet-input. $$$$ add more hashes.
        */
 
       /* Compute ingress LB hash */
-      hash_key = hm->hash_fn ((ethernet_header_t *) b0->data);
+      hash_key = hm->hash_fn ((ethernet_header_t *)
+			      vlib_buffer_get_current (b[0]));
       hash = (u32) clib_xxhash (hash_key);
 
       /* if input node did not specify next index, then packet
          should go to eternet-input */
-      if (PREDICT_FALSE ((b0->flags & VNET_BUFFER_F_HANDOFF_NEXT_VALID) == 0))
-	vnet_buffer (b0)->handoff.next_index =
-	  HANDOFF_DISPATCH_NEXT_ETHERNET_INPUT;
-      else if (vnet_buffer (b0)->handoff.next_index ==
-	       HANDOFF_DISPATCH_NEXT_IP4_INPUT
-	       || vnet_buffer (b0)->handoff.next_index ==
-	       HANDOFF_DISPATCH_NEXT_IP6_INPUT
-	       || vnet_buffer (b0)->handoff.next_index ==
-	       HANDOFF_DISPATCH_NEXT_MPLS_INPUT)
-	vlib_buffer_advance (b0, (sizeof (ethernet_header_t)));
 
       if (PREDICT_TRUE (is_pow2 (vec_len (ihd0->workers))))
 	index0 = hash & (vec_len (ihd0->workers) - 1);
       else
 	index0 = hash % vec_len (ihd0->workers);
 
-      next_worker_index += ihd0->workers[index0];
-
-      if (next_worker_index != current_worker_index)
-	{
-	  if (hf)
-	    hf->n_vectors = VLIB_FRAME_SIZE - n_left_to_next_worker;
-
-	  hf = vlib_get_worker_handoff_queue_elt (hm->frame_queue_index,
-						  next_worker_index,
-						  handoff_queue_elt_by_worker_index);
-
-	  n_left_to_next_worker = VLIB_FRAME_SIZE - hf->n_vectors;
-	  to_next_worker = &hf->buffer_index[hf->n_vectors];
-	  current_worker_index = next_worker_index;
-	}
-
-      /* enqueue to correct worker thread */
-      to_next_worker[0] = bi0;
-      to_next_worker++;
-      n_left_to_next_worker--;
-
-      if (n_left_to_next_worker == 0)
-	{
-	  hf->n_vectors = VLIB_FRAME_SIZE;
-	  vlib_put_frame_queue_elt (hf);
-	  current_worker_index = ~0;
-	  handoff_queue_elt_by_worker_index[next_worker_index] = 0;
-	  hf = 0;
-	}
+      ti[0] = hm->first_worker_index + ihd0->workers[index0];
 
       if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE)
-			 && (b0->flags & VLIB_BUFFER_IS_TRACED)))
+			 && (b[0]->flags & VLIB_BUFFER_IS_TRACED)))
 	{
 	  worker_handoff_trace_t *t =
-	    vlib_add_trace (vm, node, b0, sizeof (*t));
+	    vlib_add_trace (vm, node, b[0], sizeof (*t));
 	  t->sw_if_index = sw_if_index0;
-	  t->next_worker_index = next_worker_index - hm->first_worker_index;
-	  t->buffer_index = bi0;
+	  t->next_worker_index = ti[0];
+	  t->buffer_index = vlib_get_buffer_index (vm, b[0]);
 	}
 
+      /* next */
+      n_left_from -= 1;
+      ti += 1;
+      b += 1;
     }
 
-  if (hf)
-    hf->n_vectors = VLIB_FRAME_SIZE - n_left_to_next_worker;
-
-  /* Ship frames to the worker nodes */
-  for (i = 0; i < vec_len (handoff_queue_elt_by_worker_index); i++)
-    {
-      if (handoff_queue_elt_by_worker_index[i])
-	{
-	  hf = handoff_queue_elt_by_worker_index[i];
-	  /*
-	   * It works better to let the handoff node
-	   * rate-adapt, always ship the handoff queue element.
-	   */
-	  if (1 || hf->n_vectors == hf->last_n_vectors)
-	    {
-	      vlib_put_frame_queue_elt (hf);
-	      handoff_queue_elt_by_worker_index[i] = 0;
-	    }
-	  else
-	    hf->last_n_vectors = hf->n_vectors;
-	}
-      congested_handoff_queue_by_worker_index[i] =
-	(vlib_frame_queue_t *) (~0);
-    }
-  hf = 0;
-  current_worker_index = ~0;
+  vlib_buffer_enqueue_to_thread (vm, hm->frame_queue_index, from,
+				 thread_indices, frame->n_vectors);
   return frame->n_vectors;
 }
 
@@ -255,8 +173,10 @@ interface_handoff_enable_disable (vlib_main_t * vm, u32 sw_if_index,
     return VNET_API_ERROR_INVALID_WORKER;
 
   if (hm->frame_queue_index == ~0)
-    hm->frame_queue_index =
-      vlib_frame_queue_main_init (handoff_dispatch_node.index, 0);
+    {
+      vlib_node_t *n = vlib_get_node_by_name (vm, (u8 *) "ethernet-input");
+      hm->frame_queue_index = vlib_frame_queue_main_init (n->index, 0);
+    }
 
   vec_validate (hm->if_data, sw_if_index);
   d = vec_elt_at_index (hm->if_data, sw_if_index);
@@ -359,196 +279,6 @@ VLIB_CLI_COMMAND (set_interface_handoff_command, static) = {
 };
 /* *INDENT-ON* */
 
-typedef struct
-{
-  u32 buffer_index;
-  u32 next_index;
-  u32 sw_if_index;
-} handoff_dispatch_trace_t;
-
-/* packet trace format function */
-static u8 *
-format_handoff_dispatch_trace (u8 * s, va_list * args)
-{
-  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
-  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
-  handoff_dispatch_trace_t *t = va_arg (*args, handoff_dispatch_trace_t *);
-
-  s = format (s, "handoff-dispatch: sw_if_index %d next_index %d buffer 0x%x",
-	      t->sw_if_index, t->next_index, t->buffer_index);
-  return s;
-}
-
-#define foreach_handoff_dispatch_error \
-_(EXAMPLE, "example packets")
-
-typedef enum
-{
-#define _(sym,str) HANDOFF_DISPATCH_ERROR_##sym,
-  foreach_handoff_dispatch_error
-#undef _
-    HANDOFF_DISPATCH_N_ERROR,
-} handoff_dispatch_error_t;
-
-static char *handoff_dispatch_error_strings[] = {
-#define _(sym,string) string,
-  foreach_handoff_dispatch_error
-#undef _
-};
-
-static uword
-handoff_dispatch_node_fn (vlib_main_t * vm,
-			  vlib_node_runtime_t * node, vlib_frame_t * frame)
-{
-  u32 n_left_from, *from, *to_next;
-  handoff_dispatch_next_t next_index;
-
-  from = vlib_frame_vector_args (frame);
-  n_left_from = frame->n_vectors;
-  next_index = node->cached_next_index;
-
-  while (n_left_from > 0)
-    {
-      u32 n_left_to_next;
-
-      vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
-
-      while (n_left_from >= 4 && n_left_to_next >= 2)
-	{
-	  u32 bi0, bi1;
-	  vlib_buffer_t *b0, *b1;
-	  u32 next0, next1;
-	  u32 sw_if_index0, sw_if_index1;
-
-	  /* Prefetch next iteration. */
-	  {
-	    vlib_buffer_t *p2, *p3;
-
-	    p2 = vlib_get_buffer (vm, from[2]);
-	    p3 = vlib_get_buffer (vm, from[3]);
-
-	    vlib_prefetch_buffer_header (p2, LOAD);
-	    vlib_prefetch_buffer_header (p3, LOAD);
-	  }
-
-	  /* speculatively enqueue b0 and b1 to the current next frame */
-	  to_next[0] = bi0 = from[0];
-	  to_next[1] = bi1 = from[1];
-	  from += 2;
-	  to_next += 2;
-	  n_left_from -= 2;
-	  n_left_to_next -= 2;
-
-	  b0 = vlib_get_buffer (vm, bi0);
-	  b1 = vlib_get_buffer (vm, bi1);
-
-	  next0 = vnet_buffer (b0)->handoff.next_index;
-	  next1 = vnet_buffer (b1)->handoff.next_index;
-
-	  if (PREDICT_FALSE (vm->trace_main.trace_active_hint))
-	    {
-	      if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
-		{
-		  vlib_trace_buffer (vm, node, next0, b0,	/* follow_chain */
-				     0);
-		  handoff_dispatch_trace_t *t =
-		    vlib_add_trace (vm, node, b0, sizeof (*t));
-		  sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
-		  t->sw_if_index = sw_if_index0;
-		  t->next_index = next0;
-		  t->buffer_index = bi0;
-		}
-	      if (PREDICT_FALSE (b1->flags & VLIB_BUFFER_IS_TRACED))
-		{
-		  vlib_trace_buffer (vm, node, next1, b1,	/* follow_chain */
-				     0);
-		  handoff_dispatch_trace_t *t =
-		    vlib_add_trace (vm, node, b1, sizeof (*t));
-		  sw_if_index1 = vnet_buffer (b1)->sw_if_index[VLIB_RX];
-		  t->sw_if_index = sw_if_index1;
-		  t->next_index = next1;
-		  t->buffer_index = bi1;
-		}
-	    }
-
-	  /* verify speculative enqueues, maybe switch current next frame */
-	  vlib_validate_buffer_enqueue_x2 (vm, node, next_index,
-					   to_next, n_left_to_next,
-					   bi0, bi1, next0, next1);
-	}
-
-      while (n_left_from > 0 && n_left_to_next > 0)
-	{
-	  u32 bi0;
-	  vlib_buffer_t *b0;
-	  u32 next0;
-	  u32 sw_if_index0;
-
-	  /* speculatively enqueue b0 to the current next frame */
-	  bi0 = from[0];
-	  to_next[0] = bi0;
-	  from += 1;
-	  to_next += 1;
-	  n_left_from -= 1;
-	  n_left_to_next -= 1;
-
-	  b0 = vlib_get_buffer (vm, bi0);
-
-	  next0 = vnet_buffer (b0)->handoff.next_index;
-
-	  if (PREDICT_FALSE (vm->trace_main.trace_active_hint))
-	    {
-	      if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
-		{
-		  vlib_trace_buffer (vm, node, next0, b0,	/* follow_chain */
-				     0);
-		  handoff_dispatch_trace_t *t =
-		    vlib_add_trace (vm, node, b0, sizeof (*t));
-		  sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
-		  t->sw_if_index = sw_if_index0;
-		  t->next_index = next0;
-		  t->buffer_index = bi0;
-		}
-	    }
-
-	  /* verify speculative enqueue, maybe switch current next frame */
-	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
-					   to_next, n_left_to_next,
-					   bi0, next0);
-	}
-
-      vlib_put_next_frame (vm, node, next_index, n_left_to_next);
-    }
-
-  return frame->n_vectors;
-}
-
-/* *INDENT-OFF* */
-VLIB_REGISTER_NODE (handoff_dispatch_node) = {
-  .function = handoff_dispatch_node_fn,
-  .name = "handoff-dispatch",
-  .vector_size = sizeof (u32),
-  .format_trace = format_handoff_dispatch_trace,
-  .type = VLIB_NODE_TYPE_INTERNAL,
-  .flags = VLIB_NODE_FLAG_IS_HANDOFF,
-
-  .n_errors = ARRAY_LEN(handoff_dispatch_error_strings),
-  .error_strings = handoff_dispatch_error_strings,
-
-  .n_next_nodes = HANDOFF_DISPATCH_N_NEXT,
-
-  .next_nodes = {
-        [HANDOFF_DISPATCH_NEXT_DROP] = "error-drop",
-        [HANDOFF_DISPATCH_NEXT_ETHERNET_INPUT] = "ethernet-input",
-        [HANDOFF_DISPATCH_NEXT_IP4_INPUT] = "ip4-input-no-checksum",
-        [HANDOFF_DISPATCH_NEXT_IP6_INPUT] = "ip6-input",
-        [HANDOFF_DISPATCH_NEXT_MPLS_INPUT] = "mpls-input",
-  },
-};
-
-VLIB_NODE_FUNCTION_MULTIARCH (handoff_dispatch_node, handoff_dispatch_node_fn)
-/* *INDENT-ON* */
-
 clib_error_t *
 handoff_init (vlib_main_t * vm)
 {
@@ -574,10 +304,6 @@ handoff_init (vlib_main_t * vm)
     }
 
   hm->hash_fn = eth_get_key;
-
-  hm->vlib_main = vm;
-  hm->vnet_main = &vnet_main;
-
   hm->frame_queue_index = ~0;
 
   return 0;
diff --git a/src/vnet/handoff.h b/src/vnet/handoff.h
index 04ba8bfbc02..f50b86d5c6d 100644
--- a/src/vnet/handoff.h
+++ b/src/vnet/handoff.h
@@ -22,17 +22,6 @@
 #include <vnet/ip/ip6_packet.h>
 #include <vnet/mpls/packet.h>
 
-typedef enum
-{
-  HANDOFF_DISPATCH_NEXT_IP4_INPUT,
-  HANDOFF_DISPATCH_NEXT_IP6_INPUT,
-  HANDOFF_DISPATCH_NEXT_MPLS_INPUT,
-  HANDOFF_DISPATCH_NEXT_ETHERNET_INPUT,
-  HANDOFF_DISPATCH_NEXT_DROP,
-  HANDOFF_DISPATCH_N_NEXT,
-} handoff_dispatch_next_t;
-
-
 static inline u64
 ipv4_get_key (ip4_header_t * ip)
 {
diff --git a/src/vnet/pg/input.c b/src/vnet/pg/input.c
index a9510b7c9c9..82c6c81cf18 100644
--- a/src/vnet/pg/input.c
+++ b/src/vnet/pg/input.c
@@ -1520,9 +1520,6 @@ pg_generate_packets (vlib_node_runtime_t * node,
 	  {
 	    vlib_buffer_t *b;
 	    b = vlib_get_buffer (vm, to_next[i]);
-	    vnet_buffer (b)->device_input_feat.saved_next_index =
-	      s->next_index;
-	    vnet_buffer (b)->device_input_feat.buffer_advance = 0;
 	    b->current_config_index = current_config_index;
 	    vnet_buffer (b)->feature_arc_index = feature_arc_index;
 	  }