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>

3 files changed, 13 insertions, 14 deletions

diff --git a/src/plugins/tlsmbedtls/tls_mbedtls.c b/src/plugins/tlsmbedtls/tls_mbedtls.c
index e1262981cc2..6b56f5f0604 100644
--- a/src/plugins/tlsmbedtls/tls_mbedtls.c
+++ b/src/plugins/tlsmbedtls/tls_mbedtls.c
@@ -443,12 +443,12 @@ mbedtls_ctx_write (tls_ctx_t * ctx, session_t * app_session)
 
   ASSERT (mc->ssl.state == MBEDTLS_SSL_HANDSHAKE_OVER);
 
-  deq_max = svm_fifo_max_dequeue (app_session->tx_fifo);
+  deq_max = svm_fifo_max_dequeue_cons (app_session->tx_fifo);
   if (!deq_max)
     return 0;
 
   tls_session = session_get_from_handle (ctx->tls_session_handle);
-  enq_max = svm_fifo_max_enqueue (tls_session->tx_fifo);
+  enq_max = svm_fifo_max_enqueue_prod (tls_session->tx_fifo);
   deq_now = clib_min (deq_max, TLS_CHUNK_SIZE);
 
   if (PREDICT_FALSE (enq_max == 0))
@@ -493,12 +493,12 @@ mbedtls_ctx_read (tls_ctx_t * ctx, session_t * tls_session)
       return 0;
     }
 
-  deq_max = svm_fifo_max_dequeue (tls_session->rx_fifo);
+  deq_max = svm_fifo_max_dequeue_cons (tls_session->rx_fifo);
   if (!deq_max)
     return 0;
 
   app_session = session_get_from_handle (ctx->app_session_handle);
-  enq_max = svm_fifo_max_enqueue (app_session->rx_fifo);
+  enq_max = svm_fifo_max_enqueue_prod (app_session->rx_fifo);
   enq_now = clib_min (enq_max, TLS_CHUNK_SIZE);
 
   if (PREDICT_FALSE (enq_now == 0))
@@ -520,7 +520,7 @@ mbedtls_ctx_read (tls_ctx_t * ctx, session_t * tls_session)
   ASSERT (enq == read);
   vec_reset_length (mm->rx_bufs[thread_index]);
 
-  if (svm_fifo_max_dequeue (tls_session->rx_fifo))
+  if (svm_fifo_max_dequeue_cons (tls_session->rx_fifo))
     tls_add_vpp_q_builtin_rx_evt (tls_session);
 
   if (enq > 0)
diff --git a/src/plugins/tlsopenssl/tls_openssl.c b/src/plugins/tlsopenssl/tls_openssl.c
index e41d50a2b8b..d99a89c3d9c 100644
--- a/src/plugins/tlsopenssl/tls_openssl.c
+++ b/src/plugins/tlsopenssl/tls_openssl.c
@@ -112,7 +112,7 @@ openssl_try_handshake_read (openssl_ctx_t * oc, session_t * tls_session)
   int wrote, rv;
 
   f = tls_session->rx_fifo;
-  deq_max = svm_fifo_max_dequeue (f);
+  deq_max = svm_fifo_max_dequeue_cons (f);
   if (!deq_max)
     return 0;
 
@@ -146,7 +146,7 @@ openssl_try_handshake_write (openssl_ctx_t * oc, session_t * tls_session)
     return 0;
 
   f = tls_session->tx_fifo;
-  enq_max = svm_fifo_max_enqueue (f);
+  enq_max = svm_fifo_max_enqueue_prod (f);
   if (!enq_max)
     return 0;
 
@@ -306,7 +306,7 @@ openssl_ctx_write (tls_ctx_t * ctx, session_t * app_session)
   svm_fifo_t *f;
 
   f = app_session->tx_fifo;
-  deq_max = svm_fifo_max_dequeue (f);
+  deq_max = svm_fifo_max_dequeue_cons (f);
   if (!deq_max)
     goto check_tls_fifo;
 
@@ -342,7 +342,7 @@ check_tls_fifo:
 
   tls_session = session_get_from_handle (ctx->tls_session_handle);
   f = tls_session->tx_fifo;
-  enq_max = svm_fifo_max_enqueue (f);
+  enq_max = svm_fifo_max_enqueue_prod (f);
   if (!enq_max)
     {
       tls_add_vpp_q_builtin_tx_evt (app_session);
@@ -390,7 +390,7 @@ openssl_ctx_read (tls_ctx_t * ctx, session_t * tls_session)
     }
 
   f = tls_session->rx_fifo;
-  deq_max = svm_fifo_max_dequeue (f);
+  deq_max = svm_fifo_max_dequeue_cons (f);
   max_space = max_buf - BIO_ctrl_pending (oc->wbio);
   max_space = max_space < 0 ? 0 : max_space;
   deq_now = clib_min (deq_max, max_space);
@@ -415,7 +415,7 @@ openssl_ctx_read (tls_ctx_t * ctx, session_t * tls_session)
 	  wrote += rv;
 	}
     }
-  if (svm_fifo_max_dequeue (f))
+  if (svm_fifo_max_dequeue_cons (f))
     tls_add_vpp_q_builtin_rx_evt (tls_session);
 
 check_app_fifo:
@@ -425,7 +425,7 @@ check_app_fifo:
 
   app_session = session_get_from_handle (ctx->app_session_handle);
   f = app_session->rx_fifo;
-  enq_max = svm_fifo_max_enqueue (f);
+  enq_max = svm_fifo_max_enqueue_prod (f);
   if (!enq_max)
     {
       tls_add_vpp_q_builtin_rx_evt (tls_session);
diff --git a/src/plugins/unittest/tcp_test.c b/src/plugins/unittest/tcp_test.c
index fb01adb6c1f..c8d9d5fe247 100644
--- a/src/plugins/unittest/tcp_test.c
+++ b/src/plugins/unittest/tcp_test.c
@@ -1221,8 +1221,7 @@ tcp_test_fifo3 (vlib_main_t * vm, unformat_input_t * input)
 
   /* manually set head and tail pointers to validate modular arithmetic */
   fifo_initial_offset = fifo_initial_offset % fifo_size;
-  f->head = fifo_initial_offset;
-  f->tail = fifo_initial_offset;
+  svm_fifo_init_pointers (f, fifo_initial_offset);
 
   for (i = !randomize; i < vec_len (generate); i++)
     {