Move a part of pthread_create to native code. NFC (#14858)

src/library_pthread.js

@@ -164,9 +164,6 @@ var LibraryPThread = {
     freeThreadData: function(pthread) {
       if (!pthread) return;
       if (pthread.threadInfoStruct) {
-        var tlsMemory = {{{ makeGetValue('pthread.threadInfoStruct', C_STRUCTS.pthread.tsd, 'i32') }}};
-        {{{ makeSetValue('pthread.threadInfoStruct', C_STRUCTS.pthread.tsd, 0, 'i32') }}};
-        _free(tlsMemory);
 #if PTHREADS_PROFILING
         var profilerBlock = {{{ makeGetValue('pthread.threadInfoStruct', C_STRUCTS.pthread.profilerBlock, 'i32') }}};
         {{{ makeSetValue('pthread.threadInfoStruct',  C_STRUCTS.pthread.profilerBlock, 0, 'i32') }}};
@@ -479,7 +476,6 @@ var LibraryPThread = {
     pthread.worker.postMessage({ 'cmd': 'cancel' });
   },
 
-  $spawnThread__deps: ['$zeroMemory'],
   $spawnThread: function(threadParams) {
     if (ENVIRONMENT_IS_PTHREAD) throw 'Internal Error! spawnThread() can only ever be called from main application thread!';
 
@@ -493,10 +489,6 @@ var LibraryPThread = {
     if (!threadParams.pthread_ptr) throw 'Internal error, no pthread ptr!';
     PThread.runningWorkers.push(worker);
 
-    // Allocate memory for thread-local storage and initialize it to zero.
-    var tlsMemory = _malloc({{{ cDefine('PTHREAD_KEYS_MAX') * 4 }}});
-    zeroMemory(tlsMemory, {{{ cDefine('PTHREAD_KEYS_MAX') * 4 }}});
-
     var stackHigh = threadParams.stackBase + threadParams.stackSize;
 
     // Create a pthread info object to represent this thread.
@@ -512,18 +504,12 @@ var LibraryPThread = {
     // spawnThread is always called with a zero-initialized thread struct so
     // no need to set any valudes to zero here.
     Atomics.store(HEAPU32, tis + ({{{ C_STRUCTS.pthread.detached }}} >> 2), threadParams.detached);
-    Atomics.store(HEAPU32, tis + ({{{ C_STRUCTS.pthread.tsd }}} >> 2), tlsMemory); // Init thread-local-storage memory array.
-    Atomics.store(HEAPU32, tis + ({{{ C_STRUCTS.pthread.tid }}} >> 2), pthread.threadInfoStruct); // Main thread ID.
     Atomics.store(HEAPU32, tis + ({{{ C_STRUCTS.pthread.stack_size }}} >> 2), threadParams.stackSize);
     Atomics.store(HEAPU32, tis + ({{{ C_STRUCTS.pthread.stack }}} >> 2), stackHigh);
     Atomics.store(HEAPU32, tis + ({{{ C_STRUCTS.pthread.attr }}} >> 2), threadParams.stackSize);
     Atomics.store(HEAPU32, tis + ({{{ C_STRUCTS.pthread.attr }}} + 8 >> 2), stackHigh);
     Atomics.store(HEAPU32, tis + ({{{ C_STRUCTS.pthread.attr }}} + 12 >> 2), threadParams.detached);
 
-    var global_libc = _emscripten_get_global_libc();
-    var global_locale = global_libc + {{{ C_STRUCTS.libc.global_locale }}};
-    Atomics.store(HEAPU32, tis + ({{{ C_STRUCTS.pthread.locale }}} >> 2), global_locale);
-
 #if PTHREADS_PROFILING
     PThread.createProfilerBlock(pthread.threadInfoStruct);
 #endif
@@ -565,7 +551,7 @@ var LibraryPThread = {
 #endif
     return navigator['hardwareConcurrency'];
   },
-    
+
   __pthread_create_js__sig: 'iiiii',
   __pthread_create_js__deps: ['$spawnThread', 'pthread_self', 'memalign', 'emscripten_sync_run_in_main_thread_4'],
   __pthread_create_js: function(pthread_ptr, attr, start_routine, arg) {
@@ -573,10 +559,6 @@ var LibraryPThread = {
       err('Current environment does not support SharedArrayBuffer, pthreads are not available!');
       return {{{ cDefine('EAGAIN') }}};
     }
-    if (!pthread_ptr) {
-      err('pthread_create called with a null thread pointer!');
-      return {{{ cDefine('EINVAL') }}};
-    }
 
     // List of JS objects that will transfer ownership to the Worker hosting the thread
     var transferList = [];
@@ -729,26 +711,12 @@ var LibraryPThread = {
       assert(stackBase > 0);
     }
 
-    // Allocate thread block (pthread_t structure).
-    var threadInfoStruct = _malloc({{{ C_STRUCTS.pthread.__size__ }}});
-    // zero-initialize thread structure.
-    zeroMemory(threadInfoStruct, {{{ C_STRUCTS.pthread.__size__ }}});
-    {{{ makeSetValue('pthread_ptr', 0, 'threadInfoStruct', 'i32') }}};
-
-    // The pthread struct has a field that points to itself - this is used as a
-    // magic ID to detect whether the pthread_t structure is 'alive'.
-    {{{ makeSetValue('threadInfoStruct', C_STRUCTS.pthread.self, 'threadInfoStruct', 'i32') }}};
-
-    // pthread struct robust_list head should point to itself.
-    var headPtr = threadInfoStruct + {{{ C_STRUCTS.pthread.robust_list }}};
-    {{{ makeSetValue('headPtr', 0, 'headPtr', 'i32') }}};
-
 #if OFFSCREENCANVAS_SUPPORT
     // Register for each of the transferred canvases that the new thread now
     // owns the OffscreenCanvas.
     for (var i in offscreenCanvases) {
       // pthread ptr to the thread that owns this canvas.
-      {{{ makeSetValue('offscreenCanvases[i].canvasSharedPtr', 8, 'threadInfoStruct', 'i32') }}};
+      {{{ makeSetValue('offscreenCanvases[i].canvasSharedPtr', 8, 'pthread_ptr', 'i32') }}};
     }
 #endif
 
@@ -758,7 +726,7 @@ var LibraryPThread = {
       allocatedOwnStack: allocatedOwnStack,
       detached: detached,
       startRoutine: start_routine,
-      pthread_ptr: threadInfoStruct,
+      pthread_ptr: pthread_ptr,
       arg: arg,
 #if OFFSCREENCANVAS_SUPPORT
       moduleCanvasId: moduleCanvasId,
@@ -964,7 +932,7 @@ var LibraryPThread = {
 
   __cxa_thread_atexit__sig: 'vii',
   __cxa_thread_atexit: function(routine, arg) {
-    PThread.threadExitHandlers.push(function() { {{{ makeDynCall('vi', 'routine') }}}(arg) }); 
+    PThread.threadExitHandlers.push(function() { {{{ makeDynCall('vi', 'routine') }}}(arg) });
   },
   __cxa_thread_atexit_impl: '__cxa_thread_atexit',
 

system/lib/libc/musl/src/thread/thrd_create.c

@@ -1,7 +1,14 @@
 #include "pthread_impl.h"
 #include <threads.h>
 
+#ifdef __EMSCRIPTEN__
+// Fix for lsan.  Since lsan wraps calls to the public `pthread_create` function
+// if we call the internal __pthread_create function here to don't the wrapping
+// See pthread_create wrapper in compiler-rt/lib/lsan/lsan_interceptors.cpp.
+#define __pthread_create pthread_create
+#else
 int __pthread_create(pthread_t *restrict, const pthread_attr_t *restrict, void *(*)(void *), void *restrict);
+#endif
 
 int thrd_create(thrd_t *thr, thrd_start_t func, void *arg)
 {

system/lib/pthread/library_pthread.c

@@ -156,13 +156,12 @@ void emscripten_async_waitable_close(em_queued_call* call) {
 
 extern double emscripten_receive_on_main_thread_js(int functionIndex, int numCallArgs, double* args);
 extern int _emscripten_notify_thread_queue(pthread_t targetThreadId, pthread_t mainThreadId);
+extern int __pthread_create_js(struct pthread *thread, const pthread_attr_t *attr, void *(*start_routine) (void *), void *arg);
 
 #if defined(__has_feature)
 #if __has_feature(leak_sanitizer) || __has_feature(address_sanitizer)
 #define HAS_SANITIZER
 #include <sanitizer/lsan_interface.h>
-int emscripten_builtin_pthread_create(pthread_t *thread, const pthread_attr_t* attr,
-                                      void *(*callback)(void *), void *arg);
 #endif
 #endif
 
@@ -178,13 +177,13 @@ static void _do_call(em_queued_call* q) {
       // On non-main threads, pthread_create gets proxied to the main thread, where LSan is not
       // disabled. This makes it necessary for us to disable LSan here, so that it does not detect
       // pthread's internal allocations as leaks.
+      // If/when we remove all the allocations from __pthread_create_js we could also remove this.
       __lsan_disable();
+#endif
       q->returnValue.i =
-        emscripten_builtin_pthread_create(q->args[0].vp, q->args[1].vp, q->args[2].vp, q->args[3].vp);
+        __pthread_create_js(q->args[0].vp, q->args[1].vp, q->args[2].vp, q->args[3].vp);
+#ifdef HAS_SANITIZER
       __lsan_enable();
-#else
-      q->returnValue.i =
-        pthread_create(q->args[0].vp, q->args[1].vp, q->args[2].vp, q->args[3].vp);
 #endif
       break;
     case EM_PROXIED_CREATE_CONTEXT:
@@ -892,8 +891,8 @@ void __emscripten_pthread_data_constructor(void) {
   // The pthread struct has a field that points to itself - this is used as
   // a magic ID to detect whether the pthread_t structure is 'alive'.
   __main_pthread.self = &__main_pthread;
-      // pthread struct robust_list head should point to itself.
-  __main_pthread.robust_list.head = &__main_pthread.robust_list;
+  // pthread struct robust_list head should point to itself.
+  __main_pthread.robust_list.head = &__main_pthread.robust_list.head;
 
   // Main thread ID.
   __main_pthread.tid = (long)&__main_pthread;

system/lib/pthread/pthread_create.c

@@ -10,12 +10,14 @@
 #include "assert.h"
 #include <pthread.h>
 #include <stdbool.h>
+#include <string.h>
 #include <threads.h>
+#include <emscripten/emmalloc.h>
 
 // See musl's pthread_create.c
 
 extern int __cxa_thread_atexit(void (*)(void *), void *, void *);
-extern int __pthread_create_js(pthread_t *thread, const pthread_attr_t *attr, void *(*start_routine) (void *), void *arg);
+extern int __pthread_create_js(struct pthread *thread, const pthread_attr_t *attr, void *(*start_routine) (void *), void *arg);
 extern void _emscripten_thread_init(int, int, int);
 extern void __pthread_exit_run_handlers();
 extern void __pthread_exit_done();
@@ -51,8 +53,35 @@ void __do_cleanup_pop(struct __ptcb *cb) {
   __pthread_self()->cancelbuf = cb->__next;
 }
 
-int __pthread_create(pthread_t *thread, const pthread_attr_t *attr, void *(*start_routine) (void *), void *arg) {
-  return __pthread_create_js(thread, attr, start_routine, arg);
+int __pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict attrp, void *(*entry)(void *), void *restrict arg) {
+  // Note on LSAN: lsan intercepts/wraps calls to pthread_create so any
+  // allocation we we do here should be considered leaks.
+  // See: lsan_interceptors.cpp.
+  if (!res) {
+    return EINVAL;
+  }
+
+  // Allocate thread block (pthread_t structure).
+  struct pthread *new = malloc(sizeof(struct pthread));
+  // zero-initialize thread structure.
+  memset(new, 0, sizeof(struct pthread));
+
+  // The pthread struct has a field that points to itself - this is used as a
+  // magic ID to detect whether the pthread_t structure is 'alive'.
+  new->self = new;
+  new->tid = (uintptr_t)new;
+
+  // pthread struct robust_list head should point to itself.
+  new->robust_list.head = &new->robust_list.head;
+
+  new->locale = &libc.global_locale;
+
+  // Allocate memory for thread-local storage and initialize it to zero.
+  new->tsd = malloc(PTHREAD_KEYS_MAX * sizeof(void*));
+  memset(new->tsd, 0, PTHREAD_KEYS_MAX * sizeof(void*));
+
+  *res = new;
+  return __pthread_create_js(new, attrp, entry, arg);
 }
 
 void _emscripten_thread_exit(void* result) {
@@ -75,6 +104,11 @@ void _emscripten_thread_exit(void* result) {
     return;
   }
 
+  // We have the call the buildin free here since lsan handling for this thread
+  // gets shut down during __pthread_tsd_run_dtors.
+  emscripten_builtin_free(self->tsd);
+  self->tsd = NULL;
+
   // Mark the thread as no longer running.
   // When we publish this, the main thread is free to deallocate the thread object and we are done.
   self->threadStatus = 1;

tests/other/metadce/minimal_main_Oz_USE_PTHREADS_PROXY_TO_PTHREAD.exports

@@ -14,7 +14,6 @@ M
 N
 O
 P
-Q
 v
 w
 x

tests/other/metadce/minimal_main_Oz_USE_PTHREADS_PROXY_TO_PTHREAD.funcs

@@ -1,7 +1,6 @@
 $GetQueue
 $__emscripten_pthread_data_constructor
 $__errno_location
-$__pthread_create
 $__pthread_mutex_lock
 $__pthread_mutex_trylock
 $__pthread_mutex_unlock
@@ -27,7 +26,6 @@ $em_queued_call_free
 $em_queued_call_malloc
 $emscripten_async_run_in_main_thread
 $emscripten_current_thread_process_queued_calls
-$emscripten_get_global_libc
 $emscripten_main_thread_process_queued_calls
 $emscripten_proxy_main
 $emscripten_run_in_main_runtime_thread_js
@@ -38,6 +36,7 @@ $emscripten_tls_init
 $free_tls
 $init_mparams
 $main
+$memset
 $sbrk
 $stackAlloc
 $stackRestore