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>

3 files changed, 24 insertions, 28 deletions

diff --git a/src/vlib/unix/cj.c b/src/vlib/unix/cj.c
index 33ba163a..7c1e9475 100644
--- a/src/vlib/unix/cj.c
+++ b/src/vlib/unix/cj.c
@@ -48,7 +48,7 @@ cj_log (u32 type, void *data0, void *data1)
 
   r = (cj_record_t *) & (cjm->records[new_tail & (cjm->num_records - 1)]);
   r->time = vlib_time_now (cjm->vlib_main);
-  r->cpu = os_get_cpu_number ();
+  r->thread_index = vlib_get_thread_index ();
   r->type = type;
   r->data[0] = pointer_to_uword (data0);
   r->data[1] = pointer_to_uword (data1);
@@ -133,7 +133,8 @@ static inline void
 cj_dump_one_record (cj_record_t * r)
 {
   fprintf (stderr, "[%d]: %10.6f T%02d %llx %llx\n",
-	   r->cpu, r->time, r->type, (long long unsigned int) r->data[0],
+	   r->thread_index, r->time, r->type,
+	   (long long unsigned int) r->data[0],
 	   (long long unsigned int) r->data[1]);
 }
 
@@ -161,7 +162,7 @@ cj_dump_internal (u8 filter0_enable, u64 filter0,
   index = (cjm->tail + 1) & (cjm->num_records - 1);
   r = &(cjm->records[index]);
 
-  if (r->cpu != (u32) ~ 0)
+  if (r->thread_index != (u32) ~ 0)
     {
       /* Yes, dump from tail + 1 to the end */
       for (i = index; i < cjm->num_records; i++)
diff --git a/src/vlib/unix/cj.h b/src/vlib/unix/cj.h
index 67626afe..d0a1d46e 100644
--- a/src/vlib/unix/cj.h
+++ b/src/vlib/unix/cj.h
@@ -23,7 +23,7 @@
 typedef struct
 {
   f64 time;
-  u32 cpu;
+  u32 thread_index;
   u32 type;
   u64 data[2];
 } cj_record_t;
diff --git a/src/vlib/unix/main.c b/src/vlib/unix/main.c
index 6b96cc0d..db5ddd64 100644
--- a/src/vlib/unix/main.c
+++ b/src/vlib/unix/main.c
@@ -510,13 +510,28 @@ thread0 (uword arg)
   return i;
 }
 
+u8 *
+vlib_thread_stack_init (uword thread_index)
+{
+  vec_validate (vlib_thread_stacks, thread_index);
+  vlib_thread_stacks[thread_index] = clib_mem_alloc_aligned
+    (VLIB_THREAD_STACK_SIZE, VLIB_THREAD_STACK_SIZE);
+
+  /*
+   * Disallow writes to the bottom page of the stack, to
+   * catch stack overflows.
+   */
+  if (mprotect (vlib_thread_stacks[thread_index],
+		clib_mem_get_page_size (), PROT_READ) < 0)
+    clib_unix_warning ("thread stack");
+  return vlib_thread_stacks[thread_index];
+}
+
 int
 vlib_unix_main (int argc, char *argv[])
 {
   vlib_main_t *vm = &vlib_global_main;	/* one and only time for this! */
-  vlib_thread_main_t *tm = &vlib_thread_main;
   unformat_input_t input;
-  u8 *thread_stacks;
   clib_error_t *e;
   int i;
 
@@ -548,29 +563,9 @@ vlib_unix_main (int argc, char *argv[])
     }
   unformat_free (&input);
 
-  /*
-   * allocate n x VLIB_THREAD_STACK_SIZE stacks, aligned to a
-   * VLIB_THREAD_STACK_SIZE boundary
-   * See also: os_get_cpu_number() in vlib/vlib/threads.c
-   */
-  thread_stacks = clib_mem_alloc_aligned
-    ((uword) tm->n_thread_stacks * VLIB_THREAD_STACK_SIZE,
-     VLIB_THREAD_STACK_SIZE);
-
-  vec_validate (vlib_thread_stacks, tm->n_thread_stacks - 1);
-  for (i = 0; i < vec_len (vlib_thread_stacks); i++)
-    {
-      vlib_thread_stacks[i] = thread_stacks;
-
-      /*
-       * Disallow writes to the bottom page of the stack, to
-       * catch stack overflows.
-       */
-      if (mprotect (thread_stacks, clib_mem_get_page_size (), PROT_READ) < 0)
-	clib_unix_warning ("thread stack");
+  vlib_thread_stack_init (0);
 
-      thread_stacks += VLIB_THREAD_STACK_SIZE;
-    }
+  vlib_thread_index = 0;
 
   i = clib_calljmp (thread0, (uword) vm,
 		    (void *) (vlib_thread_stacks[0] +