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>

4 files changed, 21 insertions, 21 deletions

diff --git a/src/plugins/lb/lb.c b/src/plugins/lb/lb.c
index add81236ae5..addc2a42d34 100644
--- a/src/plugins/lb/lb.c
+++ b/src/plugins/lb/lb.c
@@ -63,11 +63,11 @@ u8 *format_lb_main (u8 * s, va_list * args)
   s = format(s, " #vips: %u\n", pool_elts(lbm->vips));
   s = format(s, " #ass: %u\n", pool_elts(lbm->ass) - 1);
 
-  u32 cpu_index;
-  for(cpu_index = 0; cpu_index < tm->n_vlib_mains; cpu_index++ ) {
-    lb_hash_t *h = lbm->per_cpu[cpu_index].sticky_ht;
+  u32 thread_index;
+  for(thread_index = 0; thread_index < tm->n_vlib_mains; thread_index++ ) {
+    lb_hash_t *h = lbm->per_cpu[thread_index].sticky_ht;
     if (h) {
-      s = format(s, "core %d\n", cpu_index);
+      s = format(s, "core %d\n", thread_index);
       s = format(s, "  timeout: %ds\n", h->timeout);
       s = format(s, "  usage: %d / %d\n", lb_hash_elts(h, lb_hash_time_now(vlib_get_main())),  lb_hash_size(h));
     }
diff --git a/src/plugins/lb/node.c b/src/plugins/lb/node.c
index 8b763c537d5..3171148bd3d 100644
--- a/src/plugins/lb/node.c
+++ b/src/plugins/lb/node.c
@@ -60,10 +60,10 @@ format_lb_trace (u8 * s, va_list * args)
   return s;
 }
 
-lb_hash_t *lb_get_sticky_table(u32 cpu_index)
+lb_hash_t *lb_get_sticky_table(u32 thread_index)
 {
   lb_main_t *lbm = &lb_main;
-  lb_hash_t *sticky_ht = lbm->per_cpu[cpu_index].sticky_ht;
+  lb_hash_t *sticky_ht = lbm->per_cpu[thread_index].sticky_ht;
   //Check if size changed
   if (PREDICT_FALSE(sticky_ht && (lbm->per_cpu_sticky_buckets != lb_hash_nbuckets(sticky_ht))))
     {
@@ -71,8 +71,8 @@ lb_hash_t *lb_get_sticky_table(u32 cpu_index)
       lb_hash_bucket_t *b;
       u32 i;
       lb_hash_foreach_entry(sticky_ht, b, i) {
-	vlib_refcount_add(&lbm->as_refcount, cpu_index, b->value[i], -1);
-	vlib_refcount_add(&lbm->as_refcount, cpu_index, 0, 1);
+	vlib_refcount_add(&lbm->as_refcount, thread_index, b->value[i], -1);
+	vlib_refcount_add(&lbm->as_refcount, thread_index, 0, 1);
       }
 
       lb_hash_free(sticky_ht);
@@ -81,8 +81,8 @@ lb_hash_t *lb_get_sticky_table(u32 cpu_index)
 
   //Create if necessary
   if (PREDICT_FALSE(sticky_ht == NULL)) {
-    lbm->per_cpu[cpu_index].sticky_ht = lb_hash_alloc(lbm->per_cpu_sticky_buckets, lbm->flow_timeout);
-    sticky_ht = lbm->per_cpu[cpu_index].sticky_ht;
+    lbm->per_cpu[thread_index].sticky_ht = lb_hash_alloc(lbm->per_cpu_sticky_buckets, lbm->flow_timeout);
+    sticky_ht = lbm->per_cpu[thread_index].sticky_ht;
     clib_warning("Regenerated sticky table %p", sticky_ht);
   }
 
@@ -153,10 +153,10 @@ lb_node_fn (vlib_main_t * vm,
 {
   lb_main_t *lbm = &lb_main;
   u32 n_left_from, *from, next_index, *to_next, n_left_to_next;
-  u32 cpu_index = os_get_cpu_number();
+  u32 thread_index = vlib_get_thread_index();
   u32 lb_time = lb_hash_time_now(vm);
 
-  lb_hash_t *sticky_ht = lb_get_sticky_table(cpu_index);
+  lb_hash_t *sticky_ht = lb_get_sticky_table(thread_index);
   from = vlib_frame_vector_args (frame);
   n_left_from = frame->n_vectors;
   next_index = node->cached_next_index;
@@ -240,9 +240,9 @@ lb_node_fn (vlib_main_t * vm,
 	  //Configuration may be changed, vectors resized, etc...
 
 	  //Dereference previously used
-	  vlib_refcount_add(&lbm->as_refcount, cpu_index,
+	  vlib_refcount_add(&lbm->as_refcount, thread_index,
 			    lb_hash_available_value(sticky_ht, hash0, available_index0), -1);
-	  vlib_refcount_add(&lbm->as_refcount, cpu_index,
+	  vlib_refcount_add(&lbm->as_refcount, thread_index,
 			    asindex0, 1);
 
 	  //Add sticky entry
@@ -260,7 +260,7 @@ lb_node_fn (vlib_main_t * vm,
 	}
 
       vlib_increment_simple_counter(&lbm->vip_counters[counter],
-				    cpu_index,
+				    thread_index,
 				    vnet_buffer (p0)->ip.adj_index[VLIB_TX],
 				    1);
 
diff --git a/src/plugins/lb/refcount.c b/src/plugins/lb/refcount.c
index 22415c8889e..6f01ab5aaf7 100644
--- a/src/plugins/lb/refcount.c
+++ b/src/plugins/lb/refcount.c
@@ -31,10 +31,10 @@ u64 vlib_refcount_get(vlib_refcount_t *r, u32 index)
 {
   u64 count = 0;
   vlib_thread_main_t *tm = vlib_get_thread_main ();
-  u32 cpu_index;
-  for (cpu_index = 0; cpu_index < tm->n_vlib_mains; cpu_index++) {
-    if (r->per_cpu[cpu_index].length > index)
-      count += r->per_cpu[cpu_index].counters[index];
+  u32 thread_index;
+  for (thread_index = 0; thread_index < tm->n_vlib_mains; thread_index++) {
+    if (r->per_cpu[thread_index].length > index)
+      count += r->per_cpu[thread_index].counters[index];
   }
   return count;
 }
diff --git a/src/plugins/lb/refcount.h b/src/plugins/lb/refcount.h
index 8c26e7be76f..dcfcb3fee58 100644
--- a/src/plugins/lb/refcount.h
+++ b/src/plugins/lb/refcount.h
@@ -45,9 +45,9 @@ typedef struct {
 void __vlib_refcount_resize(vlib_refcount_per_cpu_t *per_cpu, u32 size);
 
 static_always_inline
-void vlib_refcount_add(vlib_refcount_t *r, u32 cpu_index, u32 counter_index, i32 v)
+void vlib_refcount_add(vlib_refcount_t *r, u32 thread_index, u32 counter_index, i32 v)
 {
-  vlib_refcount_per_cpu_t *per_cpu = &r->per_cpu[cpu_index];
+  vlib_refcount_per_cpu_t *per_cpu = &r->per_cpu[thread_index];
   if (PREDICT_FALSE(counter_index >= per_cpu->length))
     __vlib_refcount_resize(per_cpu, clib_max(counter_index + 16, per_cpu->length * 2));