interface: add multi tx-queues support for new tx infra

Type: feature

Change-Id: I231f782b3c56dc2b10321e4569ac7acdad1c11da
Signed-off-by: Mohsin Kazmi <sykazmi@cisco.com>
diff --git a/src/vnet/interface/runtime.c b/src/vnet/interface/runtime.c
index e63f1ec..5c215e8 100644
--- a/src/vnet/interface/runtime.c
+++ b/src/vnet/interface/runtime.c
@@ -184,39 +184,73 @@
 	}
     }
 
-  new_out_runtimes =
-    vec_dup_aligned (hi->output_node_thread_runtimes, CLIB_CACHE_LINE_BYTES);
-  vec_validate_aligned (new_out_runtimes, n_threads - 1,
-			CLIB_CACHE_LINE_BYTES);
-
-  if (vec_len (hi->output_node_thread_runtimes) != vec_len (new_out_runtimes))
-    something_changed_on_tx = 1;
-
-  for (int i = 0; i < vec_len (hi->tx_queue_indices); i++)
+  if (vec_len (hi->tx_queue_indices) > 0)
     {
-      u32 thread_index;
-      u32 queue_index = hi->tx_queue_indices[i];
-      vnet_hw_if_tx_queue_t *txq = vnet_hw_if_get_tx_queue (vnm, queue_index);
-      uword n_threads = clib_bitmap_count_set_bits (txq->threads);
+      new_out_runtimes = vec_dup_aligned (hi->output_node_thread_runtimes,
+					  CLIB_CACHE_LINE_BYTES);
+      vec_validate_aligned (new_out_runtimes, n_threads - 1,
+			    CLIB_CACHE_LINE_BYTES);
 
-      clib_bitmap_foreach (thread_index, txq->threads)
+      for (u32 i = 0; i < vec_len (new_out_runtimes); i++)
 	{
 	  vnet_hw_if_output_node_runtime_t *rt;
-	  rt = vec_elt_at_index (new_out_runtimes, thread_index);
-	  if ((rt->frame.queue_id != txq->queue_id) ||
-	      (rt->n_threads != n_threads))
+	  rt = vec_elt_at_index (new_out_runtimes, i);
+	  u32 n_queues = 0, total_queues = vec_len (hi->tx_queue_indices);
+	  rt->frame = 0;
+	  rt->lookup_table = 0;
+
+	  for (u32 j = 0; j < total_queues; j++)
 	    {
+	      u32 queue_index = hi->tx_queue_indices[j];
+	      vnet_hw_if_tx_frame_t frame = { .shared_queue = 0,
+					      .hints = 7,
+					      .queue_id = ~0 };
+	      vnet_hw_if_tx_queue_t *txq =
+		vnet_hw_if_get_tx_queue (vnm, queue_index);
+	      if (!clib_bitmap_get (txq->threads, i))
+		continue;
+
 	      log_debug ("tx queue data changed for interface %v, thread %u "
-			 "(queue_id %u -> %u, n_threads %u -> %u)",
-			 hi->name, thread_index, rt->frame.queue_id,
-			 txq->queue_id, rt->n_threads, n_threads);
+			 "(queue_id %u)",
+			 hi->name, i, txq->queue_id);
 	      something_changed_on_tx = 1;
-	      rt->frame.queue_id = txq->queue_id;
-	      rt->frame.shared_queue = txq->shared_queue;
-	      rt->n_threads = n_threads;
+
+	      frame.queue_id = txq->queue_id;
+	      frame.shared_queue = txq->shared_queue;
+	      vec_add1 (rt->frame, frame);
+	      n_queues++;
+	    }
+
+	  // don't initialize rt->n_queues above
+	  if (rt->n_queues != n_queues)
+	    {
+	      something_changed_on_tx = 1;
+	      rt->n_queues = n_queues;
+	    }
+	  /*
+	   * It is only used in case of multiple txq.
+	   */
+	  if (rt->n_queues > 0)
+	    {
+	      if (!is_pow2 (n_queues))
+		n_queues = max_pow2 (n_queues);
+
+	      vec_validate_aligned (rt->lookup_table, n_queues - 1,
+				    CLIB_CACHE_LINE_BYTES);
+
+	      for (u32 k = 0; k < vec_len (rt->lookup_table); k++)
+		{
+		  rt->lookup_table[k] = rt->frame[k % rt->n_queues].queue_id;
+		  log_debug ("tx queue lookup table changed for interface %v, "
+			     "(lookup table [%u]=%u)",
+			     hi->name, k, rt->lookup_table[k]);
+		}
 	    }
 	}
     }
+  else
+    /* interface deleted */
+    something_changed_on_tx = 1;
 
   if (something_changed_on_rx || something_changed_on_tx)
     {
@@ -303,6 +337,11 @@
     {
       vec_free (d[i]);
       vec_free (a[i]);
+      if (new_out_runtimes)
+	{
+	  vec_free (new_out_runtimes[i].frame);
+	  vec_free (new_out_runtimes[i].lookup_table);
+	}
     }
 
   vec_free (d);