Avoid proxying atexit calls back to main thread.

To achieve this we manage the `atexit` functions in native code using existing musl code. This is small step towards a large change to just use musl for all `atexit` handling: #14479. The codesize implications of this change are a mixed bag. In some places we see saving but in other cases the extra export causes a small regression (only when EXIT_RUNTIME=1). In the long, once we land #14479 there should be more code size saving to be had by doing everything on the native side. Fixes #15868
emscripten-core · Jan 7, 2022 · e89185d · e89185d
1 parent b77adc9
commit e89185d
Show file tree

Hide file tree

Showing 32 changed files with 82 additions and 36 deletions.
diff --git a/emcc.py b/emcc.py
@@ -2328,6 +2328,9 @@ def check_memory_setting(setting):
     # enables the --post-emscripten pass
     settings.GLOBAL_BASE = 1024
 
+  if settings.EXIT_RUNTIME:
+    settings.REQUIRED_EXPORTS += ['__funcs_on_exit']
+
   if settings.MINIMAL_RUNTIME:
     if settings.EXIT_RUNTIME:
       settings.DEFAULT_LIBRARY_FUNCS_TO_INCLUDE += ['proc_exit', '$callRuntimeCallbacks']

diff --git a/src/library.js b/src/library.js
@@ -361,21 +361,6 @@ LibraryManager.library = {
   // stdlib.h
   // ==========================================================================
 
-#if MINIMAL_RUNTIME && !EXIT_RUNTIME
-  atexit__sig: 'v', // atexit unsupported in MINIMAL_RUNTIME
-  atexit: function(){},
-  __cxa_atexit: function(){},
-#else
-  atexit__proxy: 'sync',
-  atexit__sig: 'iii',
-  atexit: function(func, arg) {
-#if EXIT_RUNTIME
-    __ATEXIT__.unshift({ func: func, arg: arg });
-#endif
-  },
-  __cxa_atexit: 'atexit',
-#endif
-
   // TODO: There are currently two abort() functions that get imported to asm
   // module scope: the built-in runtime function abort(), and this library
   // function _abort(). Remove one of these, importing two functions for the

diff --git a/src/preamble.js b/src/preamble.js
@@ -430,6 +430,9 @@ function exitRuntime() {
   if (ENVIRONMENT_IS_PTHREAD) return; // PThreads reuse the runtime from the main thread.
 #endif
 #if EXIT_RUNTIME
+#if !STANDALONE_WASM
+  ___funcs_on_exit(); // Native atexit() functions
+#endif
   callRuntimeCallbacks(__ATEXIT__);
   <<< ATEXITS >>>
 #endif

diff --git a/system/lib/libc/musl/src/exit/atexit.c b/system/lib/libc/musl/src/exit/atexit.c
@@ -49,6 +49,11 @@ int __cxa_atexit(void (*func)(void *), void *arg, void *dso)
 
 	/* If the current function list is full, add a new one */
 	if (slot==COUNT) {
+#ifdef __EMSCRIPTEN__
+		// Avoid the malloc dependency and just abort if we exceed the
+		// the static limit.
+		abort();
+#else
 		struct fl *new_fl = calloc(sizeof(struct fl), 1);
 		if (!new_fl) {
 			UNLOCK(lock);
@@ -57,6 +62,7 @@ int __cxa_atexit(void (*func)(void *), void *arg, void *dso)
 		new_fl->next = head;
 		head = new_fl;
 		slot = 0;
+#endif
 	}
 
 	/* Append function to the list. */

diff --git a/tests/other/metadce/hello_libcxx_O2.jssize b/tests/other/metadce/hello_libcxx_O2.jssize
@@ -1 +1 @@
-98361
+98236
diff --git a/tests/other/metadce/hello_libcxx_O2.sent b/tests/other/metadce/hello_libcxx_O2.sent
@@ -1,4 +1,3 @@
-__cxa_atexit
 abort
 emscripten_memcpy_big
 emscripten_resize_heap

diff --git a/tests/other/metadce/hello_libcxx_O2.size b/tests/other/metadce/hello_libcxx_O2.size
@@ -1 +1 @@
-124705
+124869
diff --git a/tests/other/metadce/hello_libcxx_O2_fexceptions.jssize b/tests/other/metadce/hello_libcxx_O2_fexceptions.jssize
@@ -1 +1 @@
-111833
+111708
diff --git a/tests/other/metadce/hello_libcxx_O2_fexceptions.sent b/tests/other/metadce/hello_libcxx_O2_fexceptions.sent
@@ -1,5 +1,4 @@
 __cxa_allocate_exception
-__cxa_atexit
 __cxa_begin_catch
 __cxa_end_catch
 __cxa_find_matching_catch_2

diff --git a/tests/other/metadce/hello_libcxx_O2_fexceptions.size b/tests/other/metadce/hello_libcxx_O2_fexceptions.size
@@ -1 +1 @@
-166697
+166882
diff --git a/tests/other/metadce/hello_libcxx_O2_fexceptions_DEMANGLE_SUPPORT.jssize b/tests/other/metadce/hello_libcxx_O2_fexceptions_DEMANGLE_SUPPORT.jssize
@@ -1 +1 @@
-112820
+112695
diff --git a/tests/other/metadce/hello_libcxx_O2_fexceptions_DEMANGLE_SUPPORT.sent b/tests/other/metadce/hello_libcxx_O2_fexceptions_DEMANGLE_SUPPORT.sent
@@ -1,5 +1,4 @@
 __cxa_allocate_exception
-__cxa_atexit
 __cxa_begin_catch
 __cxa_end_catch
 __cxa_find_matching_catch_2

diff --git a/tests/other/metadce/hello_libcxx_O2_fexceptions_DEMANGLE_SUPPORT.size b/tests/other/metadce/hello_libcxx_O2_fexceptions_DEMANGLE_SUPPORT.size
@@ -1 +1 @@
-226724
+226909
diff --git a/tests/other/metadce/libcxxabi_message_O3_STANDALONE_WASM.exports b/tests/other/metadce/libcxxabi_message_O3_STANDALONE_WASM.exports
@@ -1,3 +1,4 @@
+__funcs_on_exit
 __indirect_function_table
 _start
 memory
diff --git a/tests/other/metadce/libcxxabi_message_O3_STANDALONE_WASM.jssize b/tests/other/metadce/libcxxabi_message_O3_STANDALONE_WASM.jssize
@@ -1 +1 @@
-15003
+15208
diff --git a/tests/other/metadce/libcxxabi_message_O3_STANDALONE_WASM.size b/tests/other/metadce/libcxxabi_message_O3_STANDALONE_WASM.size
@@ -1 +1 @@
-181
+199
diff --git a/tests/other/metadce/mem_O3_ALLOW_MEMORY_GROWTH_STANDALONE_WASM.exports b/tests/other/metadce/mem_O3_ALLOW_MEMORY_GROWTH_STANDALONE_WASM.exports
@@ -1,3 +1,4 @@
+__funcs_on_exit
 __indirect_function_table
 _start
 memory
diff --git a/tests/other/metadce/mem_O3_ALLOW_MEMORY_GROWTH_STANDALONE_WASM.jssize b/tests/other/metadce/mem_O3_ALLOW_MEMORY_GROWTH_STANDALONE_WASM.jssize
@@ -1 +1 @@
-15898
+16103
diff --git a/tests/other/metadce/mem_O3_ALLOW_MEMORY_GROWTH_STANDALONE_WASM.size b/tests/other/metadce/mem_O3_ALLOW_MEMORY_GROWTH_STANDALONE_WASM.size
@@ -1 +1 @@
-6303
+6321
diff --git a/tests/other/metadce/mem_O3_STANDALONE_WASM.exports b/tests/other/metadce/mem_O3_STANDALONE_WASM.exports
@@ -1,3 +1,4 @@
+__funcs_on_exit
 __indirect_function_table
 _start
 memory
diff --git a/tests/other/metadce/mem_O3_STANDALONE_WASM.jssize b/tests/other/metadce/mem_O3_STANDALONE_WASM.jssize
@@ -1 +1 @@
-15720
+15925
diff --git a/tests/other/metadce/mem_O3_STANDALONE_WASM.size b/tests/other/metadce/mem_O3_STANDALONE_WASM.size
@@ -1 +1 @@
-6226
+6244
diff --git a/tests/other/metadce/mem_no_argv_O3_STANDALONE_WASM.exports b/tests/other/metadce/mem_no_argv_O3_STANDALONE_WASM.exports
@@ -1,3 +1,4 @@
+__funcs_on_exit
 __indirect_function_table
 _start
 memory
diff --git a/tests/other/metadce/mem_no_argv_O3_STANDALONE_WASM.jssize b/tests/other/metadce/mem_no_argv_O3_STANDALONE_WASM.jssize
@@ -1 +1 @@
-15003
+15208
diff --git a/tests/other/metadce/mem_no_argv_O3_STANDALONE_WASM.size b/tests/other/metadce/mem_no_argv_O3_STANDALONE_WASM.size
@@ -1 +1 @@
-6026
+6044
diff --git a/tests/other/metadce/mem_no_argv_O3_STANDALONE_WASM_flto.exports b/tests/other/metadce/mem_no_argv_O3_STANDALONE_WASM_flto.exports
@@ -1,3 +1,4 @@
+__funcs_on_exit
 __indirect_function_table
 _start
 memory
diff --git a/tests/other/metadce/mem_no_argv_O3_STANDALONE_WASM_flto.jssize b/tests/other/metadce/mem_no_argv_O3_STANDALONE_WASM_flto.jssize
@@ -1 +1 @@
-15003
+15208
diff --git a/tests/other/metadce/mem_no_argv_O3_STANDALONE_WASM_flto.size b/tests/other/metadce/mem_no_argv_O3_STANDALONE_WASM_flto.size
@@ -1 +1 @@
-4707
+4725
diff --git a/tests/pthread/test_pthread_busy_wait_atexit.cpp b/tests/pthread/test_pthread_busy_wait_atexit.cpp
@@ -0,0 +1,37 @@
+#include <stdatomic.h>
+#include <stdbool.h>
+#include <pthread.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <emscripten/console.h>
+
+_Atomic bool done = false;
+
+void exit_handler() {
+  printf("exit_handler\n");
+}
+
+void* thread_main(void*) {
+  // Avoid using printf here since stdio is proxies back to the
+  // main thread which is busy looping
+  _emscripten_out("in thread");
+  atexit(exit_handler);
+  done = true;
+  return NULL;
+}
+
+// Similar to test_pthread_busy_wait.cpp but with lower level pthreads
+// API and explcit use of atexit before setting done to true.
+// We also don't make any calls during the busy loop which means that
+// proxied calls are *not* processed.
+int main() {
+  printf("in main\n");
+  pthread_t t;
+  pthread_create(&t, NULL, thread_main, NULL);
+
+  while (!done) { }
+
+  pthread_join(t, NULL);
+  printf("done main\n");
+  return 0;
+}
diff --git a/tests/pthread/test_pthread_busy_wait_atexit.out b/tests/pthread/test_pthread_busy_wait_atexit.out
@@ -0,0 +1,4 @@
+in main
+in thread
+done main
+exit_handler
diff --git a/tests/test_core.py b/tests/test_core.py
@@ -8486,6 +8486,12 @@ def test_pthread_busy_wait(self):
     self.set_setting('EXIT_RUNTIME')
     self.do_run_in_out_file_test('pthread/test_pthread_busy_wait.cpp')
 
+  @node_pthreads
+  def test_pthread_busy_wait_atexit(self):
+    self.set_setting('PTHREAD_POOL_SIZE', 1)
+    self.set_setting('EXIT_RUNTIME')
+    self.do_run_in_out_file_test('pthread/test_pthread_busy_wait_atexit.cpp')
+
   @node_pthreads
   def test_pthread_create_pool(self):
     # with a pool, we can synchronously depend on workers being available

diff --git a/tools/system_libs.py b/tools/system_libs.py
@@ -918,7 +918,7 @@ def get_files(self):
 
     libc_files += files_in_path(
         path='system/lib/libc/musl/src/exit',
-        filenames=['_Exit.c'])
+        filenames=['_Exit.c', 'atexit.c'])
 
     libc_files += files_in_path(
         path='system/lib/libc/musl/src/ldso',
@@ -1562,7 +1562,7 @@ def get_files(self):
     # including fprintf etc.
     files += files_in_path(
         path='system/lib/libc/musl/src/exit',
-        filenames=['assert.c', 'atexit.c', 'exit.c'])
+        filenames=['assert.c', 'exit.c'])
     return files
 
   def can_use(self):