Revert 'Workaround TSan false positives in extend_size_map'

This reverts commit e522d6d791680b55825b5f11a4c082eb4770ecf5.

Because there is again a data race here, not a false positive,
though it is again not likely to fail in practice.

1 files changed, 2 insertions, 4 deletions

diff --git a/alloc.c b/alloc.c
index d9b61a6e..43a7d24f 100644
--- a/alloc.c
+++ b/alloc.c
@@ -32,13 +32,12 @@
  * The call GC_allocobj(i,k) ensures that the freelist for
  * kind k objects of size i points to a non-empty
  * free list. It returns a pointer to the first entry on the free list.
- * If not using thread-local allocation, GC_allocobj may be called to
- * allocate an object of small size lb (and NORMAL kind) as follows
+ * In a single-threaded world, GC_allocobj may be called to allocate
+ * an object of small size lb (and NORMAL kind) as follows
  * (GC_generic_malloc_inner is a wrapper over GC_allocobj which also
  * fills in GC_size_map if needed):
  *
  *   lg = GC_size_map[lb];
- *   LOCK();
  *   op = GC_objfreelist[lg];
  *   if (NULL == op) {
  *     op = GC_generic_malloc_inner(lb, NORMAL);
@@ -46,7 +45,6 @@
  *     GC_objfreelist[lg] = obj_link(op);
  *     GC_bytes_allocd += GRANULES_TO_BYTES((word)lg);
  *   }
- *   UNLOCK();
  *
  * Note that this is very fast if the free list is non-empty; it should
  * only involve the execution of 4 or 5 simple instructions.