diff --git a/system/lib/pthread/pthread_create.c b/system/lib/pthread/pthread_create.c
index 3d4b2f7b053c0..99402679122be 100644
--- a/system/lib/pthread/pthread_create.c
+++ b/system/lib/pthread/pthread_create.c
@@ -32,6 +32,8 @@
 
 static void dummy_0() {}
 weak_alias(dummy_0, __pthread_tsd_run_dtors);
+weak_alias(dummy_0, __do_orphaned_stdio_locks);
+weak_alias(dummy_0, __dl_thread_cleanup);
 
 static void __run_cleanup_handlers() {
   pthread_t self = __pthread_self();
@@ -312,6 +314,16 @@ void _emscripten_thread_exit(void* result) {
   // Call into the musl function that runs destructors of all thread-specific data.
   __pthread_tsd_run_dtors();
 
+  // If this is the main runtime thread, don't proceed with
+  // termination of the thread, but prepare for exit to call
+  // atexit handlers.
+  if (emscripten_is_main_runtime_thread()) {
+    exit(0);
+  }
+
+  /* At this point we are committed to thread termination. */
+
+  /* The thread list lock must be AS-safe. */
   __tl_lock();
 
   /* Process robust list in userspace to handle non-pshared mutexes
@@ -333,24 +345,23 @@ void _emscripten_thread_exit(void* result) {
   }
   __vm_unlock();
 
-  if (!--libc.threads_minus_1) libc.need_locks = 0;
+  __do_orphaned_stdio_locks();
+  __dl_thread_cleanup();
 
+  /* Last, unlink thread from the list. This change will not be visible
+   * until the lock is released via __tl_unlock() below. */
+  if (!--libc.threads_minus_1) libc.need_locks = 0;
   self->next->prev = self->prev;
   self->prev->next = self->next;
   self->prev = self->next = self;
 
   __tl_unlock();
 
-  if (emscripten_is_main_runtime_thread()) {
-    exit(0);
-    return;
-  }
-
   // Not hosting a pthread anymore in this worker set __pthread_self to NULL
   __set_thread_state(NULL, 0, 0, 1);
 
-  // This atomic potentially competes with a concurrent pthread_detach
-  // call; the loser is responsible for freeing thread resources.
+  /* This atomic potentially competes with a concurrent pthread_detach
+   * call; the loser is responsible for freeing thread resources. */
   int state = a_cas(&self->detach_state, DT_JOINABLE, DT_EXITING);
 
   if (state == DT_DETACHED) {
@@ -363,8 +374,10 @@ void _emscripten_thread_exit(void* result) {
     // When dynamic linking is enabled we need to keep track of zombie threads
     _emscripten_thread_exit_joinable(self);
 #endif
+
+    /* Wake any joiner. */
     a_store(&self->detach_state, DT_EXITED);
-    __wake(&self->detach_state, 1, 1); // Wake any joiner.
+    __wake(&self->detach_state, 1, 1);
   }
 }
 
diff --git a/test/codesize/test_codesize_minimal_pthreads.json b/test/codesize/test_codesize_minimal_pthreads.json
index 4e0869c3b49ed..928806850fcd3 100644
--- a/test/codesize/test_codesize_minimal_pthreads.json
+++ b/test/codesize/test_codesize_minimal_pthreads.json
@@ -2,9 +2,9 @@
   "a.out.js": 7367,
   "a.out.js.gz": 3587,
   "a.out.nodebug.wasm": 19037,
-  "a.out.nodebug.wasm.gz": 8786,
+  "a.out.nodebug.wasm.gz": 8787,
   "total": 26404,
-  "total_gz": 12373,
+  "total_gz": 12374,
   "sent": [
     "a (memory)",
     "b (exit)",
diff --git a/test/codesize/test_codesize_minimal_pthreads_memgrowth.json b/test/codesize/test_codesize_minimal_pthreads_memgrowth.json
index d4906b590fa64..5477e9ba79e30 100644
--- a/test/codesize/test_codesize_minimal_pthreads_memgrowth.json
+++ b/test/codesize/test_codesize_minimal_pthreads_memgrowth.json
@@ -2,9 +2,9 @@
   "a.out.js": 7776,
   "a.out.js.gz": 3791,
   "a.out.nodebug.wasm": 19038,
-  "a.out.nodebug.wasm.gz": 8787,
+  "a.out.nodebug.wasm.gz": 8788,
   "total": 26814,
-  "total_gz": 12578,
+  "total_gz": 12579,
   "sent": [
     "a (memory)",
     "b (exit)",