svm_fifo rework to avoid contention on cursize

Problems Addressed:
- Contention of cursize by producer and consumer.
- Reduce the no of modulo operations.

Changes:
- Synchronization between producer and consumer changed from cursize
  to head and tail indexes
  Implications: reduces the usable size of fifo by 1.
- Using weaker memory ordering C++11 atomics to access head and tail
  based on producer and consumer role.
- Head and tail indexes are unsigned 32 bit integers. Additions and
  subtraction on them are implicit 32 bit Modulo operation.
- Adding weaker memory ordering variants of max_enq, max_deq, is_empty
  and is_full Using them appropriately in all places.

Perfomance improvement (iperf3 via Hoststack):

iperf3 Server: Marvell ThunderX2(AArch64) - iperf3 Client: Skylake(x86)
   ~6%(256 rxd/txd) - ~11%(2048 rxd/txd)

Change-Id: I1d484e000e437430fdd5a819657d1c6b62443018
Signed-off-by: Sirshak Das <sirshak.das@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>

5 files changed, 18 insertions, 25 deletions

diff --git a/src/vnet/session/application_interface.h b/src/vnet/session/application_interface.h
index 8323e72e768..b49744c0b29 100644
--- a/src/vnet/session/application_interface.h
+++ b/src/vnet/session/application_interface.h
@@ -459,7 +459,7 @@ app_send_dgram_raw (svm_fifo_t * f, app_session_transport_t * at,
   session_dgram_hdr_t hdr;
   int rv;
 
-  max_enqueue = svm_fifo_max_enqueue (f);
+  max_enqueue = svm_fifo_max_enqueue_prod (f);
   if (max_enqueue <= sizeof (session_dgram_hdr_t))
     return 0;
 
@@ -533,7 +533,7 @@ app_recv_dgram_raw (svm_fifo_t * f, u8 * buf, u32 len,
   u32 max_deq;
   int rv;
 
-  max_deq = svm_fifo_max_dequeue (f);
+  max_deq = svm_fifo_max_dequeue_cons (f);
   if (max_deq < sizeof (session_dgram_hdr_t))
     {
       if (clear_evt)
diff --git a/src/vnet/session/application_worker.c b/src/vnet/session/application_worker.c
index a8edad3edbb..dd1088e2cbb 100644
--- a/src/vnet/session/application_worker.c
+++ b/src/vnet/session/application_worker.c
@@ -369,21 +369,12 @@ app_worker_own_session (app_worker_t * app_wrk, session_t * s)
   if (app_worker_alloc_session_fifos (sm, s))
     return -1;
 
-  if (!svm_fifo_is_empty (rxf))
-    {
-      clib_memcpy_fast (s->rx_fifo->data, rxf->data, rxf->nitems);
-      s->rx_fifo->head = rxf->head;
-      s->rx_fifo->tail = rxf->tail;
-      s->rx_fifo->cursize = rxf->cursize;
-    }
+  if (!svm_fifo_is_empty_cons (rxf))
+    svm_fifo_clone (s->rx_fifo, rxf);
+
+  if (!svm_fifo_is_empty_cons (txf))
+    svm_fifo_clone (s->tx_fifo, txf);
 
-  if (!svm_fifo_is_empty (txf))
-    {
-      clib_memcpy_fast (s->tx_fifo->data, txf->data, txf->nitems);
-      s->tx_fifo->head = txf->head;
-      s->tx_fifo->tail = txf->tail;
-      s->tx_fifo->cursize = txf->cursize;
-    }
 
   segment_manager_dealloc_fifos (rxf, txf);
 
diff --git a/src/vnet/session/session.c b/src/vnet/session/session.c
index 90248b64401..58ec1c01369 100644
--- a/src/vnet/session/session.c
+++ b/src/vnet/session/session.c
@@ -411,7 +411,7 @@ session_enqueue_dgram_connection (session_t * s,
 {
   int enqueued = 0, rv, in_order_off;
 
-  ASSERT (svm_fifo_max_enqueue (s->rx_fifo)
+  ASSERT (svm_fifo_max_enqueue_prod (s->rx_fifo)
 	  >= b->current_length + sizeof (*hdr));
 
   svm_fifo_enqueue_nowait (s->rx_fifo, sizeof (session_dgram_hdr_t),
@@ -508,7 +508,7 @@ session_enqueue_notify_inline (session_t * s)
   /* *INDENT-OFF* */
   SESSION_EVT_DBG(SESSION_EVT_ENQ, s, ({
       ed->data[0] = SESSION_IO_EVT_RX;
-      ed->data[1] = svm_fifo_max_dequeue (s->rx_fifo);
+      ed->data[1] = svm_fifo_max_dequeue_prod (s->rx_fifo);
   }));
   /* *INDENT-ON* */
 
@@ -1112,7 +1112,7 @@ session_transport_close (session_t * s)
    * point, either after sending everything or after a timeout, call delete
    * notify. This will finally lead to the complete cleanup of the session.
    */
-  if (svm_fifo_max_dequeue (s->tx_fifo))
+  if (svm_fifo_max_dequeue_cons (s->tx_fifo))
     s->session_state = SESSION_STATE_CLOSED_WAITING;
   else
     s->session_state = SESSION_STATE_CLOSED;
diff --git a/src/vnet/session/session.h b/src/vnet/session/session.h
index ed42e5476e9..997d15395ef 100644
--- a/src/vnet/session/session.h
+++ b/src/vnet/session/session.h
@@ -378,14 +378,14 @@ always_inline u32
 transport_max_rx_enqueue (transport_connection_t * tc)
 {
   session_t *s = session_get (tc->s_index, tc->thread_index);
-  return svm_fifo_max_enqueue (s->rx_fifo);
+  return svm_fifo_max_enqueue_prod (s->rx_fifo);
 }
 
 always_inline u32
 transport_max_tx_dequeue (transport_connection_t * tc)
 {
   session_t *s = session_get (tc->s_index, tc->thread_index);
-  return svm_fifo_max_dequeue (s->tx_fifo);
+  return svm_fifo_max_dequeue_cons (s->tx_fifo);
 }
 
 always_inline u32
diff --git a/src/vnet/session/session_node.c b/src/vnet/session/session_node.c
index 2eac5152e16..c894c437cde 100644
--- a/src/vnet/session/session_node.c
+++ b/src/vnet/session/session_node.c
@@ -82,7 +82,8 @@ session_mq_accepted_reply_handler (void *data)
     {
       old_state = s->session_state;
       s->session_state = SESSION_STATE_READY;
-      if (!svm_fifo_is_empty (s->rx_fifo))
+
+      if (!svm_fifo_is_empty_prod (s->rx_fifo))
 	app_worker_lock_and_send_event (app_wrk, s, SESSION_IO_EVT_RX);
 
       /* Closed while waiting for app to reply. Resend disconnect */
@@ -557,7 +558,7 @@ session_tx_set_dequeue_params (vlib_main_t * vm, session_tx_context_t * ctx,
 			       u32 max_segs, u8 peek_data)
 {
   u32 n_bytes_per_buf, n_bytes_per_seg;
-  ctx->max_dequeue = svm_fifo_max_dequeue (ctx->s->tx_fifo);
+  ctx->max_dequeue = svm_fifo_max_dequeue_cons (ctx->s->tx_fifo);
   if (peek_data)
     {
       /* Offset in rx fifo from where to peek data */
@@ -788,7 +789,7 @@ session_tx_fifo_read_and_snd_i (vlib_main_t * vm, vlib_node_runtime_t * node,
 	svm_fifo_overwrite_head (ctx->s->tx_fifo, (u8 *) & ctx->hdr,
 				 sizeof (session_dgram_pre_hdr_t));
       /* More data needs to be read */
-      else if (svm_fifo_max_dequeue (ctx->s->tx_fifo) > 0)
+      else if (svm_fifo_max_dequeue_cons (ctx->s->tx_fifo) > 0)
 	if (svm_fifo_set_event (ctx->s->tx_fifo))
 	  vec_add1 (wrk->pending_event_vector, *e);
     }
@@ -956,7 +957,8 @@ session_queue_node_fn (vlib_main_t * vm, vlib_node_runtime_t * node,
 	   * and the tx queue is still not empty, try to wait for some
 	   * dispatch cycles */
 	  if (!e->postponed
-	      || (e->postponed < 200 && svm_fifo_max_dequeue (s->tx_fifo)))
+	      || (e->postponed < 200
+		  && svm_fifo_max_dequeue_cons (s->tx_fifo)))
 	    {
 	      e->postponed += 1;
 	      vec_add1 (wrk->pending_disconnects, *e);