[libc] add _malloc_thread_cleanup option

Author: SchrodingerZhu

libc/config/config.json

@@ -89,6 +89,10 @@
     "LIBC_CONF_RWLOCK_DEFAULT_SPIN_COUNT": {
       "value": 100,
       "doc": "Default number of spins before blocking if a rwlock is in contention (default to 100)."
+    },
+    "LIBC_CONF_ENABLE_MALLOC_THREAD_CLEANUP": {
+      "value": false,
+      "doc": "Enable the `_malloc_thread_cleanup` weak symbol. When defined, this is function is called after `__cxa` and pthread-specific dtors. On main thread, this will be called after `atexit` functions and `.fini` dtors, right before TLS tearing down. This function can be overridden by allocators to perform cleanup. Allocators can use this symbol to avoid registering thread dtors using potentially reentrant routines."
     }
   },
   "math": {

libc/docs/configure.rst

@@ -47,6 +47,7 @@ to learn about the defaults for your platform and target.
     - ``LIBC_CONF_PRINTF_FLOAT_TO_STR_USE_MEGA_LONG_DOUBLE_TABLE``: Use large table for better printf long double performance.
     - ``LIBC_CONF_PRINTF_RUNTIME_DISPATCH``: Use dynamic dispatch for the output mechanism to reduce code size.
 * **"pthread" options**
+    - ``LIBC_CONF_ENABLE_MALLOC_THREAD_CLEANUP``: Enable the `_malloc_thread_cleanup` weak symbol. When defined, this is function is called after `__cxa` and pthread-specific dtors. On main thread, this will be called after `atexit` functions and `.fini` dtors, right before TLS tearing down. This function can be overridden by allocators to perform cleanup. Allocators can use this symbol to avoid registering thread dtors using potentially reentrant routines.
     - ``LIBC_CONF_RAW_MUTEX_DEFAULT_SPIN_COUNT``: Default number of spins before blocking if a mutex is in contention (default to 100).
     - ``LIBC_CONF_RWLOCK_DEFAULT_SPIN_COUNT``: Default number of spins before blocking if a rwlock is in contention (default to 100).
     - ``LIBC_CONF_TIMEOUT_ENSURE_MONOTONICITY``: Automatically adjust timeout to CLOCK_MONOTONIC (default to true). POSIX API may require CLOCK_REALTIME, which can be unstable and leading to unexpected behavior. This option will convert the real-time timestamp to monotonic timestamp relative to the time of call.

libc/src/__support/threads/linux/CMakeLists.txt

@@ -71,6 +71,12 @@ add_header_library(
     libc.src.__support.threads.mutex_common
 )
 
+if (LIBC_CONF_ENABLE_MALLOC_THREAD_CLEANUP)
+  set(malloc_cleanup_flags -DLIBC_COPT_ENABLE_MALLOC_THREAD_CLEANUP)
+else()
+  set(malloc_cleanup_flags)
+endif()
+
 add_object_library(
   thread
   SRCS
@@ -89,6 +95,7 @@ add_object_library(
     libc.src.__support.threads.thread_common
   COMPILE_OPTIONS
     ${libc_opt_high_flag}
+    ${malloc_cleanup_flags}
     -fno-omit-frame-pointer # This allows us to sniff out the thread args from
                             # the new thread's stack reliably.
     -Wno-frame-address      # Yes, calling __builtin_return_address with a

libc/src/__support/threads/linux/thread.cpp

@@ -482,6 +482,10 @@ int Thread::get_name(cpp::StringStream &name) const {
   return 0;
 }
 
+#ifdef LIBC_COPT_ENABLE_MALLOC_THREAD_CLEANUP
+extern "C" [[gnu::weak]] void _malloc_thread_cleanup();
+#endif // LIBC_COPT_ENABLE_MALLOC_THREAD_CLEANUP
+
 void thread_exit(ThreadReturnValue retval, ThreadStyle style) {
   auto attrib = self.attrib;
 
@@ -494,6 +498,11 @@ void thread_exit(ThreadReturnValue retval, ThreadStyle style) {
   // different thread. The destructors of thread local and TSS objects should
   // be called by the thread which owns them.
   internal::call_atexit_callbacks(attrib);
+#ifdef LIBC_COPT_ENABLE_MALLOC_THREAD_CLEANUP
+  // call _malloc_thread_cleanup after the atexit callbacks
+  if (_malloc_thread_cleanup)
+    _malloc_thread_cleanup();
+#endif // LIBC_COPT_ENABLE_MALLOC_THREAD_CLEANUP
 
   uint32_t joinable_state = uint32_t(DetachState::JOINABLE);
   if (!attrib->detach_state.compare_exchange_strong(

libc/src/stdlib/CMakeLists.txt

@@ -589,6 +589,12 @@ add_header_library(
 )
 endif()
 
+if (LIBC_CONF_ENABLE_MALLOC_THREAD_CLEANUP)
+  set(malloc_cleanup_flags -DLIBC_COPT_ENABLE_MALLOC_THREAD_CLEANUP)
+else()
+  set(malloc_cleanup_flags)
+endif()
+
 add_entrypoint_object(
   atexit
   SRCS
@@ -599,6 +605,8 @@ add_entrypoint_object(
     20 # For constinit
   DEPENDS
     .exit_handler
+  COMPILE_OPTIONS
+    ${malloc_cleanup_flags}
 )
 
 add_entrypoint_object(

libc/src/stdlib/atexit.cpp

@@ -18,6 +18,10 @@ constinit ExitCallbackList atexit_callbacks;
 Mutex handler_list_mtx(false, false, false, false);
 [[gnu::weak]] extern void teardown_main_tls();
 
+namespace internal {
+[[gnu::weak]] extern void call_atexit_callbacks();
+}
+
 extern "C" {
 
 int __cxa_atexit(AtExitCallback *callback, void *payload, void *) {