diff --git a/libc/src/stdlib/CMakeLists.txt b/libc/src/stdlib/CMakeLists.txt
--- a/libc/src/stdlib/CMakeLists.txt
+++ b/libc/src/stdlib/CMakeLists.txt
@@ -276,11 +276,11 @@
     atexit.cpp
   HDRS
     atexit.h
+  COMPILE_OPTIONS
+    -O3
   DEPENDS
     libc.src.__support.CPP.vector
-    libc.src.threads.mtx_init
-    libc.src.threads.mtx_lock
-    libc.src.threads.mtx_unlock
+    libc.src.__support.threads.thread
 )
 
 add_entrypoint_object(
diff --git a/libc/src/stdlib/atexit.cpp b/libc/src/stdlib/atexit.cpp
--- a/libc/src/stdlib/atexit.cpp
+++ b/libc/src/stdlib/atexit.cpp
@@ -9,20 +9,13 @@
 #include "src/stdlib/atexit.h"
 #include "src/__support/CPP/vector.h"
 #include "src/__support/common.h"
-#include "src/threads/mtx_init.h"
-#include "src/threads/mtx_lock.h"
-#include "src/threads/mtx_unlock.h"
+#include "src/__support/threads/mutex.h"
 
 namespace __llvm_libc {
 
 namespace {
 
-mtx_t lock;
-// TODO need an easier way to use mtx_t internally, or use pthread_mutex_t
-// with PTHREAD_MUTEX_INITIALIZER when it lands.
-struct Init {
-  Init() { __llvm_libc::mtx_init(&lock, mtx_plain); }
-} init;
+Mutex handler_list_mtx(false, false, false);
 
 // TOOD should we make cpp::vector like llvm::SmallVector<T, N> where it will
 // allocate at least N before needing dynamic allocation?
@@ -33,21 +26,21 @@
 namespace internal {
 
 void call_exit_handlers() {
-  __llvm_libc::mtx_lock(&lock);
+  handler_list_mtx.lock();
   // TODO: implement rbegin() + rend() for cpp::vector
   for (int i = handlers.size() - 1; i >= 0; i--) {
-    __llvm_libc::mtx_unlock(&lock);
+    handler_list_mtx.unlock();
     handlers[i]();
-    __llvm_libc::mtx_lock(&lock);
+    handler_list_mtx.lock();
   }
 }
 
 } // namespace internal
 
 LLVM_LIBC_FUNCTION(int, atexit, (void (*function)())) {
-  __llvm_libc::mtx_lock(&lock);
+  handler_list_mtx.lock();
   handlers.push_back(function);
-  __llvm_libc::mtx_unlock(&lock);
+  handler_list_mtx.unlock();
   return 0;
 }