Use thread local storage for thread index

This patch deprecates stack-based thread identification,
Also removes requirement that thread stacks are adjacent.

Finally, possibly annoying for some folks, it renames
all occurences of cpu_index and cpu_number with thread
index. Using word "cpu" is misleading here as thread can
be migrated ti different CPU, and also it is not related
to linux cpu index.

Change-Id: I68cdaf661e701d2336fc953dcb9978d10a70f7c1
Signed-off-by: Damjan Marion <damarion@cisco.com>

14 files changed, 52 insertions, 49 deletions

diff --git a/src/plugins/dpdk/buffer.c b/src/plugins/dpdk/buffer.c
index 2765c292..c80b3fa8 100644
--- a/src/plugins/dpdk/buffer.c
+++ b/src/plugins/dpdk/buffer.c
@@ -132,7 +132,7 @@ dpdk_buffer_delete_free_list (vlib_main_t * vm, u32 free_list_index)
   u32 merge_index;
   int i;
 
-  ASSERT (os_get_cpu_number () == 0);
+  ASSERT (vlib_get_thread_index () == 0);
 
   f = vlib_buffer_get_free_list (vm, free_list_index);
 
diff --git a/src/plugins/dpdk/device/device.c b/src/plugins/dpdk/device/device.c
index 50b26689..91661246 100644
--- a/src/plugins/dpdk/device/device.c
+++ b/src/plugins/dpdk/device/device.c
@@ -243,7 +243,7 @@ static_always_inline
   ASSERT (ring->tx_tail == 0);
 
   n_retry = 16;
-  queue_id = vm->cpu_index;
+  queue_id = vm->thread_index;
 
   do
     {
@@ -266,7 +266,7 @@ static_always_inline
 	{
 	  /* no wrap, transmit in one burst */
 	  dpdk_device_hqos_per_worker_thread_t *hqos =
-	    &xd->hqos_wt[vm->cpu_index];
+	    &xd->hqos_wt[vm->thread_index];
 
 	  ASSERT (hqos->swq != NULL);
 
@@ -332,7 +332,7 @@ dpdk_buffer_recycle (vlib_main_t * vm, vlib_node_runtime_t * node,
 		     vlib_buffer_t * b, u32 bi, struct rte_mbuf **mbp)
 {
   dpdk_main_t *dm = &dpdk_main;
-  u32 my_cpu = vm->cpu_index;
+  u32 my_cpu = vm->thread_index;
   struct rte_mbuf *mb_new;
 
   if (PREDICT_FALSE (b->flags & VLIB_BUFFER_RECYCLE) == 0)
@@ -376,7 +376,7 @@ dpdk_interface_tx (vlib_main_t * vm,
   tx_ring_hdr_t *ring;
   u32 n_on_ring;
 
-  my_cpu = vm->cpu_index;
+  my_cpu = vm->thread_index;
 
   queue_id = my_cpu;
 
diff --git a/src/plugins/dpdk/device/dpdk_priv.h b/src/plugins/dpdk/device/dpdk_priv.h
index dd40ff48..52b4ca4b 100644
--- a/src/plugins/dpdk/device/dpdk_priv.h
+++ b/src/plugins/dpdk/device/dpdk_priv.h
@@ -79,7 +79,7 @@ dpdk_update_counters (dpdk_device_t * xd, f64 now)
 {
   vlib_simple_counter_main_t *cm;
   vnet_main_t *vnm = vnet_get_main ();
-  u32 my_cpu = os_get_cpu_number ();
+  u32 thread_index = vlib_get_thread_index ();
   u64 rxerrors, last_rxerrors;
 
   /* only update counters for PMD interfaces */
@@ -96,7 +96,7 @@ dpdk_update_counters (dpdk_device_t * xd, f64 now)
       cm = vec_elt_at_index (vnm->interface_main.sw_if_counters,
 			     VNET_INTERFACE_COUNTER_RX_NO_BUF);
 
-      vlib_increment_simple_counter (cm, my_cpu, xd->vlib_sw_if_index,
+      vlib_increment_simple_counter (cm, thread_index, xd->vlib_sw_if_index,
 				     xd->stats.rx_nombuf -
 				     xd->last_stats.rx_nombuf);
     }
@@ -107,7 +107,7 @@ dpdk_update_counters (dpdk_device_t * xd, f64 now)
       cm = vec_elt_at_index (vnm->interface_main.sw_if_counters,
 			     VNET_INTERFACE_COUNTER_RX_MISS);
 
-      vlib_increment_simple_counter (cm, my_cpu, xd->vlib_sw_if_index,
+      vlib_increment_simple_counter (cm, thread_index, xd->vlib_sw_if_index,
 				     xd->stats.imissed -
 				     xd->last_stats.imissed);
     }
@@ -119,7 +119,7 @@ dpdk_update_counters (dpdk_device_t * xd, f64 now)
       cm = vec_elt_at_index (vnm->interface_main.sw_if_counters,
 			     VNET_INTERFACE_COUNTER_RX_ERROR);
 
-      vlib_increment_simple_counter (cm, my_cpu, xd->vlib_sw_if_index,
+      vlib_increment_simple_counter (cm, thread_index, xd->vlib_sw_if_index,
 				     rxerrors - last_rxerrors);
     }
 
diff --git a/src/plugins/dpdk/device/init.c b/src/plugins/dpdk/device/init.c
index 538db6cb..7eaf8da7 100755
--- a/src/plugins/dpdk/device/init.c
+++ b/src/plugins/dpdk/device/init.c
@@ -324,7 +324,7 @@ dpdk_port_setup (dpdk_main_t * dm, dpdk_device_t * xd)
   int rv;
   int j;
 
-  ASSERT (os_get_cpu_number () == 0);
+  ASSERT (vlib_get_thread_index () == 0);
 
   if (xd->flags & DPDK_DEVICE_FLAG_ADMIN_UP)
     {
diff --git a/src/plugins/dpdk/device/node.c b/src/plugins/dpdk/device/node.c
index e740fd18..b10e0fad 100644
--- a/src/plugins/dpdk/device/node.c
+++ b/src/plugins/dpdk/device/node.c
@@ -283,7 +283,7 @@ dpdk_buffer_init_from_template (void *d0, void *d1, void *d2, void *d3,
  */
 static_always_inline u32
 dpdk_device_input (dpdk_main_t * dm, dpdk_device_t * xd,
-		   vlib_node_runtime_t * node, u32 cpu_index, u16 queue_id,
+		   vlib_node_runtime_t * node, u32 thread_index, u16 queue_id,
 		   int maybe_multiseg)
 {
   u32 n_buffers;
@@ -294,7 +294,7 @@ dpdk_device_input (dpdk_main_t * dm, dpdk_device_t * xd,
   uword n_rx_bytes = 0;
   u32 n_trace, trace_cnt __attribute__ ((unused));
   vlib_buffer_free_list_t *fl;
-  vlib_buffer_t *bt = vec_elt_at_index (dm->buffer_templates, cpu_index);
+  vlib_buffer_t *bt = vec_elt_at_index (dm->buffer_templates, thread_index);
 
   if ((xd->flags & DPDK_DEVICE_FLAG_ADMIN_UP) == 0)
     return 0;
@@ -306,7 +306,7 @@ dpdk_device_input (dpdk_main_t * dm, dpdk_device_t * xd,
       return 0;
     }
 
-  vec_reset_length (xd->d_trace_buffers[cpu_index]);
+  vec_reset_length (xd->d_trace_buffers[thread_index]);
   trace_cnt = n_trace = vlib_get_trace_count (vm, node);
 
   if (n_trace > 0)
@@ -318,7 +318,7 @@ dpdk_device_input (dpdk_main_t * dm, dpdk_device_t * xd,
 	{
 	  struct rte_mbuf *mb = xd->rx_vectors[queue_id][mb_index++];
 	  vlib_buffer_t *b = vlib_buffer_from_rte_mbuf (mb);
-	  vec_add1 (xd->d_trace_buffers[cpu_index],
+	  vec_add1 (xd->d_trace_buffers[thread_index],
 		    vlib_get_buffer_index (vm, b));
 	}
     }
@@ -546,20 +546,22 @@ dpdk_device_input (dpdk_main_t * dm, dpdk_device_t * xd,
       vlib_put_next_frame (vm, node, next_index, n_left_to_next);
     }
 
-  if (PREDICT_FALSE (vec_len (xd->d_trace_buffers[cpu_index]) > 0))
+  if (PREDICT_FALSE (vec_len (xd->d_trace_buffers[thread_index]) > 0))
     {
-      dpdk_rx_trace (dm, node, xd, queue_id, xd->d_trace_buffers[cpu_index],
-		     vec_len (xd->d_trace_buffers[cpu_index]));
-      vlib_set_trace_count (vm, node, n_trace -
-			    vec_len (xd->d_trace_buffers[cpu_index]));
+      dpdk_rx_trace (dm, node, xd, queue_id,
+		     xd->d_trace_buffers[thread_index],
+		     vec_len (xd->d_trace_buffers[thread_index]));
+      vlib_set_trace_count (vm, node,
+			    n_trace -
+			    vec_len (xd->d_trace_buffers[thread_index]));
     }
 
   vlib_increment_combined_counter
     (vnet_get_main ()->interface_main.combined_sw_if_counters
      + VNET_INTERFACE_COUNTER_RX,
-     cpu_index, xd->vlib_sw_if_index, mb_index, n_rx_bytes);
+     thread_index, xd->vlib_sw_if_index, mb_index, n_rx_bytes);
 
-  vnet_device_increment_rx_packets (cpu_index, mb_index);
+  vnet_device_increment_rx_packets (thread_index, mb_index);
 
   return mb_index;
 }
@@ -630,19 +632,19 @@ dpdk_input (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * f)
   dpdk_device_t *xd;
   uword n_rx_packets = 0;
   dpdk_device_and_queue_t *dq;
-  u32 cpu_index = os_get_cpu_number ();
+  u32 thread_index = vlib_get_thread_index ();
 
   /*
    * Poll all devices on this cpu for input/interrupts.
    */
   /* *INDENT-OFF* */
-  vec_foreach (dq, dm->devices_by_cpu[cpu_index])
+  vec_foreach (dq, dm->devices_by_cpu[thread_index])
     {
       xd = vec_elt_at_index(dm->devices, dq->device);
       if (xd->flags & DPDK_DEVICE_FLAG_MAYBE_MULTISEG)
-        n_rx_packets += dpdk_device_input (dm, xd, node, cpu_index, dq->queue_id, /* maybe_multiseg */ 1);
+        n_rx_packets += dpdk_device_input (dm, xd, node, thread_index, dq->queue_id, /* maybe_multiseg */ 1);
       else
-        n_rx_packets += dpdk_device_input (dm, xd, node, cpu_index, dq->queue_id, /* maybe_multiseg */ 0);
+        n_rx_packets += dpdk_device_input (dm, xd, node, thread_index, dq->queue_id, /* maybe_multiseg */ 0);
     }
   /* *INDENT-ON* */
 
diff --git a/src/plugins/dpdk/hqos/hqos.c b/src/plugins/dpdk/hqos/hqos.c
index a288fca7..8b251beb 100644
--- a/src/plugins/dpdk/hqos/hqos.c
+++ b/src/plugins/dpdk/hqos/hqos.c
@@ -397,7 +397,7 @@ static_always_inline void
 dpdk_hqos_thread_internal_hqos_dbg_bypass (vlib_main_t * vm)
 {
   dpdk_main_t *dm = &dpdk_main;
-  u32 cpu_index = vm->cpu_index;
+  u32 thread_index = vm->thread_index;
   u32 dev_pos;
 
   dev_pos = 0;
@@ -405,12 +405,12 @@ dpdk_hqos_thread_internal_hqos_dbg_bypass (vlib_main_t * vm)
     {
       vlib_worker_thread_barrier_check ();
 
-      u32 n_devs = vec_len (dm->devices_by_hqos_cpu[cpu_index]);
+      u32 n_devs = vec_len (dm->devices_by_hqos_cpu[thread_index]);
       if (dev_pos >= n_devs)
 	dev_pos = 0;
 
       dpdk_device_and_queue_t *dq =
-	vec_elt_at_index (dm->devices_by_hqos_cpu[cpu_index], dev_pos);
+	vec_elt_at_index (dm->devices_by_hqos_cpu[thread_index], dev_pos);
       dpdk_device_t *xd = vec_elt_at_index (dm->devices, dq->device);
 
       dpdk_device_hqos_per_hqos_thread_t *hqos = xd->hqos_ht;
@@ -479,7 +479,7 @@ static_always_inline void
 dpdk_hqos_thread_internal (vlib_main_t * vm)
 {
   dpdk_main_t *dm = &dpdk_main;
-  u32 cpu_index = vm->cpu_index;
+  u32 thread_index = vm->thread_index;
   u32 dev_pos;
 
   dev_pos = 0;
@@ -487,7 +487,7 @@ dpdk_hqos_thread_internal (vlib_main_t * vm)
     {
       vlib_worker_thread_barrier_check ();
 
-      u32 n_devs = vec_len (dm->devices_by_hqos_cpu[cpu_index]);
+      u32 n_devs = vec_len (dm->devices_by_hqos_cpu[thread_index]);
       if (PREDICT_FALSE (n_devs == 0))
 	{
 	  dev_pos = 0;
@@ -497,7 +497,7 @@ dpdk_hqos_thread_internal (vlib_main_t * vm)
 	dev_pos = 0;
 
       dpdk_device_and_queue_t *dq =
-	vec_elt_at_index (dm->devices_by_hqos_cpu[cpu_index], dev_pos);
+	vec_elt_at_index (dm->devices_by_hqos_cpu[thread_index], dev_pos);
       dpdk_device_t *xd = vec_elt_at_index (dm->devices, dq->device);
 
       dpdk_device_hqos_per_hqos_thread_t *hqos = xd->hqos_ht;
@@ -586,7 +586,7 @@ dpdk_hqos_thread (vlib_worker_thread_t * w)
 
   vm = vlib_get_main ();
 
-  ASSERT (vm->cpu_index == os_get_cpu_number ());
+  ASSERT (vm->thread_index == vlib_get_thread_index ());
 
   clib_time_init (&vm->clib_time);
   clib_mem_set_heap (w->thread_mheap);
@@ -595,7 +595,7 @@ dpdk_hqos_thread (vlib_worker_thread_t * w)
   while (tm->worker_thread_release == 0)
     vlib_worker_thread_barrier_check ();
 
-  if (vec_len (dm->devices_by_hqos_cpu[vm->cpu_index]) == 0)
+  if (vec_len (dm->devices_by_hqos_cpu[vm->thread_index]) == 0)
     return
       clib_error
       ("current I/O TX thread does not have any devices assigned to it");
diff --git a/src/plugins/dpdk/ipsec/cli.c b/src/plugins/dpdk/ipsec/cli.c
index cd0a6037..3ae8c9b8 100644
--- a/src/plugins/dpdk/ipsec/cli.c
+++ b/src/plugins/dpdk/ipsec/cli.c
@@ -42,8 +42,8 @@ dpdk_ipsec_show_mapping (vlib_main_t * vm, u16 detail_display)
   for (i = 0; i < tm->n_vlib_mains; i++)
     {
       uword key, data;
-      u32 cpu_index = vlib_mains[i]->cpu_index;
-      crypto_worker_main_t *cwm = &dcm->workers_main[cpu_index];
+      u32 thread_index = vlib_mains[i]->thread_index;
+      crypto_worker_main_t *cwm = &dcm->workers_main[thread_index];
       u8 *s = 0;
 
       if (skip_master)
@@ -57,7 +57,7 @@ dpdk_ipsec_show_mapping (vlib_main_t * vm, u16 detail_display)
 	  i32 last_cdev = -1;
 	  crypto_qp_data_t *qpd;
 
-	  s = format (s, "%u\t", cpu_index);
+	  s = format (s, "%u\t", thread_index);
 
 	  /* *INDENT-OFF* */
 	  vec_foreach (qpd, cwm->qp_data)
@@ -95,7 +95,7 @@ dpdk_ipsec_show_mapping (vlib_main_t * vm, u16 detail_display)
 	    cap.sym.auth.algo = p_key->auth_algo;
 	    check_algo_is_supported (&cap, auth_str);
 	    vlib_cli_output (vm, "%u\t%10s\t%15s\t%3s\t%u\t%u\n",
-			     vlib_mains[i]->cpu_index, cipher_str, auth_str,
+			     vlib_mains[i]->thread_index, cipher_str, auth_str,
 			     p_key->is_outbound ? "out" : "in",
 			     cwm->qp_data[data].dev_id,
 			     cwm->qp_data[data].qp_id);
diff --git a/src/plugins/dpdk/ipsec/crypto_node.c b/src/plugins/dpdk/ipsec/crypto_node.c
index dc3452b2..a3c45902 100644
--- a/src/plugins/dpdk/ipsec/crypto_node.c
+++ b/src/plugins/dpdk/ipsec/crypto_node.c
@@ -171,9 +171,9 @@ static uword
 dpdk_crypto_input_fn (vlib_main_t * vm, vlib_node_runtime_t * node,
 		      vlib_frame_t * frame)
 {
-  u32 cpu_index = os_get_cpu_number ();
+  u32 thread_index = vlib_get_thread_index ();
   dpdk_crypto_main_t *dcm = &dpdk_crypto_main;
-  crypto_worker_main_t *cwm = &dcm->workers_main[cpu_index];
+  crypto_worker_main_t *cwm = &dcm->workers_main[thread_index];
   crypto_qp_data_t *qpd;
   u32 n_deq = 0;
 
diff --git a/src/plugins/dpdk/ipsec/esp.h b/src/plugins/dpdk/ipsec/esp.h
index 320295b1..56f0c756 100644
--- a/src/plugins/dpdk/ipsec/esp.h
+++ b/src/plugins/dpdk/ipsec/esp.h
@@ -170,9 +170,9 @@ static_always_inline int
 create_sym_sess (ipsec_sa_t * sa, crypto_sa_session_t * sa_sess,
 		 u8 is_outbound)
 {
-  u32 cpu_index = os_get_cpu_number ();
+  u32 thread_index = vlib_get_thread_index ();
   dpdk_crypto_main_t *dcm = &dpdk_crypto_main;
-  crypto_worker_main_t *cwm = &dcm->workers_main[cpu_index];
+  crypto_worker_main_t *cwm = &dcm->workers_main[thread_index];
   struct rte_crypto_sym_xform cipher_xform = { 0 };
   struct rte_crypto_sym_xform auth_xform = { 0 };
   struct rte_crypto_sym_xform *xfs;
diff --git a/src/plugins/dpdk/ipsec/esp_decrypt.c b/src/plugins/dpdk/ipsec/esp_decrypt.c
index 286e03f8..bab76e3b 100644
--- a/src/plugins/dpdk/ipsec/esp_decrypt.c
+++ b/src/plugins/dpdk/ipsec/esp_decrypt.c
@@ -88,7 +88,7 @@ dpdk_esp_decrypt_node_fn (vlib_main_t * vm,
 {
   u32 n_left_from, *from, *to_next, next_index;
   ipsec_main_t *im = &ipsec_main;
-  u32 cpu_index = os_get_cpu_number();
+  u32 thread_index = vlib_get_thread_index();
   dpdk_crypto_main_t * dcm = &dpdk_crypto_main;
   dpdk_esp_main_t * em = &dpdk_esp_main;
   u32 i;
@@ -104,7 +104,7 @@ dpdk_esp_decrypt_node_fn (vlib_main_t * vm,
       return n_left_from;
     }
 
-  crypto_worker_main_t *cwm = vec_elt_at_index(dcm->workers_main, cpu_index);
+  crypto_worker_main_t *cwm = vec_elt_at_index(dcm->workers_main, thread_index);
   u32 n_qps = vec_len(cwm->qp_data);
   struct rte_crypto_op ** cops_to_enq[n_qps];
   u32 n_cop_qp[n_qps], * bi_to_enq[n_qps];
diff --git a/src/plugins/dpdk/ipsec/esp_encrypt.c b/src/plugins/dpdk/ipsec/esp_encrypt.c
index 5b03de73..f996d7df 100644
--- a/src/plugins/dpdk/ipsec/esp_encrypt.c
+++ b/src/plugins/dpdk/ipsec/esp_encrypt.c
@@ -93,7 +93,7 @@ dpdk_esp_encrypt_node_fn (vlib_main_t * vm,
 {
   u32 n_left_from, *from, *to_next, next_index;
   ipsec_main_t *im = &ipsec_main;
-  u32 cpu_index = os_get_cpu_number ();
+  u32 thread_index = vlib_get_thread_index ();
   dpdk_crypto_main_t *dcm = &dpdk_crypto_main;
   dpdk_esp_main_t *em = &dpdk_esp_main;
   u32 i;
@@ -111,7 +111,8 @@ dpdk_esp_encrypt_node_fn (vlib_main_t * vm,
       return n_left_from;
     }
 
-  crypto_worker_main_t *cwm = vec_elt_at_index (dcm->workers_main, cpu_index);
+  crypto_worker_main_t *cwm =
+    vec_elt_at_index (dcm->workers_main, thread_index);
   u32 n_qps = vec_len (cwm->qp_data);
   struct rte_crypto_op **cops_to_enq[n_qps];
   u32 n_cop_qp[n_qps], *bi_to_enq[n_qps];
diff --git a/src/plugins/dpdk/ipsec/ipsec.c b/src/plugins/dpdk/ipsec/ipsec.c
index b0aaaaec..5d8f4fba 100644
--- a/src/plugins/dpdk/ipsec/ipsec.c
+++ b/src/plugins/dpdk/ipsec/ipsec.c
@@ -289,7 +289,7 @@ dpdk_ipsec_process (vlib_main_t * vm, vlib_node_runtime_t * rt,
 	      if (!map)
 		{
 		  clib_warning ("unable to create hash table for worker %u",
-				vlib_mains[i]->cpu_index);
+				vlib_mains[i]->thread_index);
 		  goto error;
 		}
 	      cwm->algo_qp_map = map;
diff --git a/src/plugins/dpdk/ipsec/ipsec.h b/src/plugins/dpdk/ipsec/ipsec.h
index 28bffc80..f0f793c0 100644
--- a/src/plugins/dpdk/ipsec/ipsec.h
+++ b/src/plugins/dpdk/ipsec/ipsec.h
@@ -95,8 +95,8 @@ static_always_inline void
 crypto_alloc_cops ()
 {
   dpdk_crypto_main_t *dcm = &dpdk_crypto_main;
-  u32 cpu_index = os_get_cpu_number ();
-  crypto_worker_main_t *cwm = &dcm->workers_main[cpu_index];
+  u32 thread_index = vlib_get_thread_index ();
+  crypto_worker_main_t *cwm = &dcm->workers_main[thread_index];
   unsigned socket_id = rte_socket_id ();
   crypto_qp_data_t *qpd;
 
diff --git a/src/plugins/dpdk/main.c b/src/plugins/dpdk/main.c
index 7ee2a785..942b8b2d 100644
--- a/src/plugins/dpdk/main.c
+++ b/src/plugins/dpdk/main.c
@@ -39,7 +39,7 @@ rte_delay_us_override (unsigned us)
    * thread then do not intercept. (Must not be called from an
    * independent pthread).
    */
-  if (os_get_cpu_number () == 0)
+  if (vlib_get_thread_index () == 0)
     {
       /*
        * We're in the vlib main thread or a vlib process. Make sure