gh-128679: Redesign tracemalloc locking (#128888)

* Use TABLES_LOCK() to protect 'tracemalloc_config.tracing'.
* Hold TABLES_LOCK() longer while accessing tables.
* tracemalloc_realloc() and tracemalloc_free() no longer
  remove the trace on reentrant call.
* _PyTraceMalloc_Stop() unregisters _PyTraceMalloc_TraceRef().
* _PyTraceMalloc_GetTraces() sets the reentrant flag.
* tracemalloc_clear_traces_unlocked() sets the reentrant flag.

Include/internal/pycore_object.h

@@ -299,12 +299,6 @@ Py_ssize_t _Py_ExplicitMergeRefcount(PyObject *op, Py_ssize_t extra);
 extern int _PyType_CheckConsistency(PyTypeObject *type);
 extern int _PyDict_CheckConsistency(PyObject *mp, int check_content);
 
-/* Update the Python traceback of an object. This function must be called
-   when a memory block is reused from a free list.
-
-   Internal function called by _Py_NewReference(). */
-extern int _PyTraceMalloc_TraceRef(PyObject *op, PyRefTracerEvent event, void*);
-
 // Fast inlined version of PyType_HasFeature()
 static inline int
 _PyType_HasFeature(PyTypeObject *type, unsigned long feature) {

Include/internal/pycore_tracemalloc.h

@@ -25,7 +25,7 @@ struct _PyTraceMalloc_Config {
     } initialized;
 
     /* Is tracemalloc tracing memory allocations?
-       Variable protected by the GIL */
+       Variable protected by the TABLES_LOCK(). */
     int tracing;
 
     /* limit of the number of frames in a traceback, 1 by default.
@@ -85,14 +85,14 @@ struct _tracemalloc_runtime_state {
     size_t peak_traced_memory;
     /* Hash table used as a set to intern filenames:
        PyObject* => PyObject*.
-       Protected by the GIL */
+       Protected by the TABLES_LOCK(). */
     _Py_hashtable_t *filenames;
     /* Buffer to store a new traceback in traceback_new().
-       Protected by the GIL. */
+       Protected by the TABLES_LOCK(). */
     struct tracemalloc_traceback *traceback;
     /* Hash table used as a set to intern tracebacks:
        traceback_t* => traceback_t*
-       Protected by the GIL */
+       Protected by the TABLES_LOCK(). */
     _Py_hashtable_t *tracebacks;
     /* pointer (void*) => trace (trace_t*).
        Protected by TABLES_LOCK(). */
@@ -144,7 +144,7 @@ extern PyObject* _PyTraceMalloc_GetTraces(void);
 extern PyObject* _PyTraceMalloc_GetObjectTraceback(PyObject *obj);
 
 /* Initialize tracemalloc */
-extern int _PyTraceMalloc_Init(void);
+extern PyStatus _PyTraceMalloc_Init(void);
 
 /* Start tracemalloc */
 extern int _PyTraceMalloc_Start(int max_nframe);

Modules/_tracemalloc.c

@@ -215,18 +215,14 @@ static struct PyModuleDef module_def = {
 PyMODINIT_FUNC
 PyInit__tracemalloc(void)
 {
-    PyObject *m;
+    PyObject *mod = PyModule_Create(&module_def);
-    m = PyModule_Create(&module_def);
+    if (mod == NULL) {
-    if (m == NULL)
-        return NULL;
-#ifdef Py_GIL_DISABLED
-    PyUnstable_Module_SetGIL(m, Py_MOD_GIL_NOT_USED);
-#endif
-
-    if (_PyTraceMalloc_Init() < 0) {
-        Py_DECREF(m);
         return NULL;
     }
 
-    return m;
+#ifdef Py_GIL_DISABLED
+    PyUnstable_Module_SetGIL(mod, Py_MOD_GIL_NOT_USED);
+#endif
+
+    return mod;
 }

Python/pylifecycle.c

@@ -707,7 +707,12 @@ pycore_create_interpreter(_PyRuntimeState *runtime,
     // the settings are loaded (so that feature_flags are set) but before
     // any calls are made to obmalloc functions.
     if (_PyMem_init_obmalloc(interp) < 0) {
-        return  _PyStatus_NO_MEMORY();
+        return _PyStatus_NO_MEMORY();
+    }
+
+    status = _PyTraceMalloc_Init();
+    if (_PyStatus_EXCEPTION(status)) {
+        return status;
     }
 
     PyThreadState *tstate = _PyThreadState_New(interp,