l2: quad loop l2fwd node

Change-Id: I8510575ee1d6b7c5ccf551766e77e40ce55e644c
Signed-off-by: Damjan Marion <damarion@cisco.com>

2 files changed, 158 insertions, 28 deletions

diff --git a/vnet/vnet/l2/l2_fib.h b/vnet/vnet/l2/l2_fib.h
index d1b8534d..8e7a52e4 100644
--- a/vnet/vnet/l2/l2_fib.h
+++ b/vnet/vnet/l2/l2_fib.h
@@ -233,6 +233,83 @@ l2fib_lookup_2 (BVT (clib_bihash) * mac_table,
     }
 }
 
+static_always_inline void
+l2fib_lookup_4 (BVT (clib_bihash) * mac_table,
+		l2fib_entry_key_t * cached_key,
+		l2fib_entry_result_t * cached_result,
+		u8 * mac0,
+		u8 * mac1,
+		u8 * mac2,
+		u8 * mac3,
+		u16 bd_index0,
+		u16 bd_index1,
+		u16 bd_index2,
+		u16 bd_index3,
+		l2fib_entry_key_t * key0,
+		l2fib_entry_key_t * key1,
+		l2fib_entry_key_t * key2,
+		l2fib_entry_key_t * key3,
+		u32 * bucket0,
+		u32 * bucket1,
+		u32 * bucket2,
+		u32 * bucket3,
+		l2fib_entry_result_t * result0,
+		l2fib_entry_result_t * result1,
+		l2fib_entry_result_t * result2,
+		l2fib_entry_result_t * result3)
+{
+  /* set up key */
+  key0->raw = l2fib_make_key (mac0, bd_index0);
+  key1->raw = l2fib_make_key (mac1, bd_index1);
+  key2->raw = l2fib_make_key (mac2, bd_index2);
+  key3->raw = l2fib_make_key (mac3, bd_index3);
+
+  if ((key0->raw == cached_key->raw) && (key1->raw == cached_key->raw) &&
+      (key2->raw == cached_key->raw) && (key3->raw == cached_key->raw))
+    {
+      /* Both hit in the one-entry cache */
+      result0->raw = cached_result->raw;
+      result1->raw = cached_result->raw;
+      result2->raw = cached_result->raw;
+      result3->raw = cached_result->raw;
+      *bucket0 = ~0;
+      *bucket1 = ~0;
+      *bucket2 = ~0;
+      *bucket3 = ~0;
+
+    }
+  else
+    {
+      BVT (clib_bihash_kv) kv0, kv1, kv2, kv3;
+
+      /*
+       * Do a regular mac table lookup
+       * Interleave lookups for packet 0 and packet 1
+       */
+      kv0.key = key0->raw;
+      kv1.key = key1->raw;
+      kv2.key = key2->raw;
+      kv3.key = key3->raw;
+      kv0.value = ~0ULL;
+      kv1.value = ~0ULL;
+      kv2.value = ~0ULL;
+      kv3.value = ~0ULL;
+
+      BV (clib_bihash_search_inline) (mac_table, &kv0);
+      BV (clib_bihash_search_inline) (mac_table, &kv1);
+      BV (clib_bihash_search_inline) (mac_table, &kv2);
+      BV (clib_bihash_search_inline) (mac_table, &kv3);
+
+      result0->raw = kv0.value;
+      result1->raw = kv1.value;
+      result2->raw = kv2.value;
+      result3->raw = kv3.value;
+
+      /* Update one-entry cache */
+      cached_key->raw = key1->raw;
+      cached_result->raw = result1->raw;
+    }
+}
 
 BVT (clib_bihash) * get_mac_table (void);
      void
diff --git a/vnet/vnet/l2/l2_fwd.c b/vnet/vnet/l2/l2_fwd.c
index 586aa3d4..710a9d9e 100644
--- a/vnet/vnet/l2/l2_fwd.c
+++ b/vnet/vnet/l2/l2_fwd.c
@@ -223,49 +223,63 @@ l2fwd_node_fn (vlib_main_t * vm,
       /* get space to enqueue frame to graph node "next_index" */
       vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
 
-      while (n_left_from >= 4 && n_left_to_next >= 2)
+      while (n_left_from >= 8 && n_left_to_next >= 4)
 	{
-	  u32 bi0, bi1;
-	  vlib_buffer_t *b0, *b1;
-	  u32 next0, next1;
-	  u32 sw_if_index0, sw_if_index1;
-	  ethernet_header_t *h0, *h1;
-	  l2fib_entry_key_t key0, key1;
-	  l2fib_entry_result_t result0, result1;
-	  u32 bucket0, bucket1;
+	  u32 bi0, bi1, bi2, bi3;
+	  vlib_buffer_t *b0, *b1, *b2, *b3;
+	  u32 next0, next1, next2, next3;
+	  u32 sw_if_index0, sw_if_index1, sw_if_index2, sw_if_index3;
+	  ethernet_header_t *h0, *h1, *h2, *h3;
+	  l2fib_entry_key_t key0, key1, key2, key3;
+	  l2fib_entry_result_t result0, result1, result2, result3;
+	  u32 bucket0, bucket1, bucket2, bucket3;
 
 	  /* 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);
-
-	    CLIB_PREFETCH (p2->data, CLIB_CACHE_LINE_BYTES, STORE);
-	    CLIB_PREFETCH (p3->data, CLIB_CACHE_LINE_BYTES, STORE);
+	    vlib_buffer_t *p4, *p5, *p6, *p7;
+
+	    p4 = vlib_get_buffer (vm, from[4]);
+	    p5 = vlib_get_buffer (vm, from[5]);
+	    p6 = vlib_get_buffer (vm, from[6]);
+	    p7 = vlib_get_buffer (vm, from[7]);
+
+	    vlib_prefetch_buffer_header (p4, LOAD);
+	    vlib_prefetch_buffer_header (p5, LOAD);
+	    vlib_prefetch_buffer_header (p6, LOAD);
+	    vlib_prefetch_buffer_header (p7, LOAD);
+
+	    CLIB_PREFETCH (p4->data, CLIB_CACHE_LINE_BYTES, STORE);
+	    CLIB_PREFETCH (p5->data, CLIB_CACHE_LINE_BYTES, STORE);
+	    CLIB_PREFETCH (p6->data, CLIB_CACHE_LINE_BYTES, STORE);
+	    CLIB_PREFETCH (p7->data, CLIB_CACHE_LINE_BYTES, STORE);
 	  }
 
 	  /* speculatively enqueue b0 and b1 to the current next frame */
 	  /* bi is "buffer index", b is pointer to the buffer */
 	  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;
+	  to_next[2] = bi2 = from[2];
+	  to_next[3] = bi3 = from[3];
+	  from += 4;
+	  to_next += 4;
+	  n_left_from -= 4;
+	  n_left_to_next -= 4;
 
 	  b0 = vlib_get_buffer (vm, bi0);
 	  b1 = vlib_get_buffer (vm, bi1);
+	  b2 = vlib_get_buffer (vm, bi2);
+	  b3 = vlib_get_buffer (vm, bi3);
 
 	  /* RX interface handles */
 	  sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
 	  sw_if_index1 = vnet_buffer (b1)->sw_if_index[VLIB_RX];
+	  sw_if_index2 = vnet_buffer (b2)->sw_if_index[VLIB_RX];
+	  sw_if_index3 = vnet_buffer (b3)->sw_if_index[VLIB_RX];
 
 	  h0 = vlib_buffer_get_current (b0);
 	  h1 = vlib_buffer_get_current (b1);
+	  h2 = vlib_buffer_get_current (b2);
+	  h3 = vlib_buffer_get_current (b3);
 
 	  if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE)))
 	    {
@@ -287,27 +301,66 @@ l2fwd_node_fn (vlib_main_t * vm,
 		  clib_memcpy (t->src, h1->src_address, 6);
 		  clib_memcpy (t->dst, h1->dst_address, 6);
 		}
+	      if (b2->flags & VLIB_BUFFER_IS_TRACED)
+		{
+		  l2fwd_trace_t *t =
+		    vlib_add_trace (vm, node, b2, sizeof (*t));
+		  t->sw_if_index = sw_if_index2;
+		  t->bd_index = vnet_buffer (b2)->l2.bd_index;
+		  clib_memcpy (t->src, h2->src_address, 6);
+		  clib_memcpy (t->dst, h2->dst_address, 6);
+		}
+	      if (b3->flags & VLIB_BUFFER_IS_TRACED)
+		{
+		  l2fwd_trace_t *t =
+		    vlib_add_trace (vm, node, b3, sizeof (*t));
+		  t->sw_if_index = sw_if_index3;
+		  t->bd_index = vnet_buffer (b3)->l2.bd_index;
+		  clib_memcpy (t->src, h3->src_address, 6);
+		  clib_memcpy (t->dst, h3->dst_address, 6);
+		}
 	    }
 
 	  /* process 2 pkts */
 #ifdef COUNTERS
-	  em->counters[node_counter_base_index + L2FWD_ERROR_L2FWD] += 2;
+	  em->counters[node_counter_base_index + L2FWD_ERROR_L2FWD] += 4;
 #endif
-	  l2fib_lookup_2 (msm->mac_table, &cached_key, &cached_result, h0->dst_address, h1->dst_address, vnet_buffer (b0)->l2.bd_index, vnet_buffer (b1)->l2.bd_index, &key0,	/* not used */
+	  /* *INDENT-OFF* */
+	  l2fib_lookup_4 (msm->mac_table, &cached_key, &cached_result,
+			  h0->dst_address, h1->dst_address,
+			  h2->dst_address, h3->dst_address,
+			  vnet_buffer (b0)->l2.bd_index,
+			  vnet_buffer (b1)->l2.bd_index,
+			  vnet_buffer (b2)->l2.bd_index,
+			  vnet_buffer (b3)->l2.bd_index,
+			  &key0,	/* not used */
 			  &key1,	/* not used */
+			  &key2,	/* not used */
+			  &key3,	/* not used */
 			  &bucket0,	/* not used */
 			  &bucket1,	/* not used */
-			  &result0, &result1);
+			  &bucket2,	/* not used */
+			  &bucket3,	/* not used */
+			  &result0,
+			  &result1,
+			  &result2,
+			  &result3);
+	  /* *INDENT-ON* */
 	  l2fwd_process (vm, node, msm, em, b0, sw_if_index0, &result0,
 			 &next0);
 	  l2fwd_process (vm, node, msm, em, b1, sw_if_index1, &result1,
 			 &next1);
+	  l2fwd_process (vm, node, msm, em, b2, sw_if_index2, &result2,
+			 &next2);
+	  l2fwd_process (vm, node, msm, em, b3, sw_if_index3, &result3,
+			 &next3);
 
 	  /* verify speculative enqueues, maybe switch current next frame */
 	  /* if next0==next1==next_index then nothing special needs to be done */
-	  vlib_validate_buffer_enqueue_x2 (vm, node, next_index,
+	  vlib_validate_buffer_enqueue_x4 (vm, node, next_index,
 					   to_next, n_left_to_next,
-					   bi0, bi1, next0, next1);
+					   bi0, bi1, bi2, bi3,
+					   next0, next1, next2, next3);
 	}
 
       while (n_left_from > 0 && n_left_to_next > 0)