bpo-40512: Store pointer to interpreter state in a thread local variable

Include/internal/pycore_pystate.h

@@ -8,6 +8,7 @@ extern "C" {
 #  error "this header requires Py_BUILD_CORE define"
 #endif
 
+#include "pycore_tls.h"   /* Thread local storage */
 #include "pycore_runtime.h"   /* PyRuntimeState */
 
 
@@ -20,16 +21,14 @@ _Py_IsMainThread(void)
     return (thread == _PyRuntime.main_thread);
 }
 
-
 static inline int
 _Py_IsMainInterpreter(PyInterpreterState *interp)
 {
     /* Use directly _PyRuntime rather than tstate->interp->runtime, since
        this function is used in performance critical code path (ceval) */
-    return (interp == _PyRuntime.interpreters.main);
+    return interp == _PyRuntime.interpreters.main;
 }
 
-
 static inline const PyConfig *
 _Py_GetMainConfig(void)
 {
@@ -40,6 +39,9 @@ _Py_GetMainConfig(void)
     return _PyInterpreterState_GetConfig(interp);
 }
 
+#ifdef _Py_THREAD_LOCAL
+extern _Py_THREAD_LOCAL PyInterpreterState *_py_current_interpreter;
+#endif
 
 /* Only handle signals on the main thread of the main interpreter. */
 static inline int
@@ -107,7 +109,6 @@ _Py_EnsureFuncTstateNotNULL(const char *func, PyThreadState *tstate)
 #define _Py_EnsureTstateNotNULL(tstate) \
     _Py_EnsureFuncTstateNotNULL(__func__, tstate)
 
-
 /* Get the current interpreter state.
 
    The macro is unsafe: it does not check for error and it can return NULL.
@@ -116,14 +117,31 @@ _Py_EnsureFuncTstateNotNULL(const char *func, PyThreadState *tstate)
 
    See also _PyInterpreterState_Get()
    and _PyGILState_GetInterpreterStateUnsafe(). */
+#ifdef _Py_THREAD_LOCAL
+#ifdef Py_DEBUG
+static inline PyInterpreterState* _PyInterpreterState_GET(void) {
+    PyInterpreterState *interp = _py_current_interpreter;
+    if (interp == NULL) {
+        Py_FatalError("no current interpreter");
+    }
+    return interp;
+}
+#else
+#define _PyInterpreterState_GET() _py_current_interpreter
+#endif
+#else
 static inline PyInterpreterState* _PyInterpreterState_GET(void) {
     PyThreadState *tstate = _PyThreadState_GET();
 #ifdef Py_DEBUG
     _Py_EnsureTstateNotNULL(tstate);
 #endif
-    return tstate->interp;
+    PyInterpreterState *interp = tstate->interp;
+    if (interp == NULL) {
+        Py_FatalError("no current interpreter");
+    }
+    return interp;
 }
-
+#endif
 
 // PyThreadState functions
 

Include/internal/pycore_tls.h

@@ -0,0 +1,18 @@
+
+#ifndef Py_CORE_TLS_H
+#define Py_CORE_TLS_H
+
+#ifndef Py_BUILD_CORE
+#  error "this header requires Py_BUILD_CORE to be defined"
+#endif
+
+#ifndef _Py_THREAD_LOCAL
+#if defined(__GNUC__)
+#define _Py_THREAD_LOCAL __thread
+#endif
+#if defined(_MSC_VER)
+#define _Py_THREAD_LOCAL __declspec(thread)
+#endif
+#endif
+
+#endif /*  Py_CORE_TLS_H */

Makefile.pre.in

@@ -1577,6 +1577,7 @@ PYTHON_HEADERS= \
 		$(srcdir)/Include/internal/pycore_structseq.h \
 		$(srcdir)/Include/internal/pycore_symtable.h \
 		$(srcdir)/Include/internal/pycore_sysmodule.h \
+		$(srcdir)/Include/internal/pycore_tls.h \
 		$(srcdir)/Include/internal/pycore_traceback.h \
 		$(srcdir)/Include/internal/pycore_tuple.h \
 		$(srcdir)/Include/internal/pycore_ucnhash.h \

Misc/NEWS.d/next/Core and Builtins/2021-10-26-13-38-13.bpo-40512.ZUQPPC.rst

@@ -0,0 +1,2 @@
+A pointer to the interpreter state is stored in a thread local variable
+on platforms that support it.

Modules/_functoolsmodule.c

@@ -703,7 +703,7 @@ functools_reduce(PyObject *self, PyObject *args)
             // bpo-42536: The GC may have untracked this args tuple. Since we're
             // recycling it, make sure it's tracked again:
             if (!_PyObject_GC_IS_TRACKED(args)) {
-                _PyObject_GC_TRACK(args);
+                PyObject_GC_Track(args);
             }
         }
     }

Modules/_io/_iomodule.c

@@ -10,7 +10,7 @@
 #define PY_SSIZE_T_CLEAN
 #include "Python.h"
 #include "_iomodule.h"
-#include "pycore_pystate.h"       // _PyInterpreterState_GET()
+#include "pycore_pystate.h"       // PyInterpreterState_Get()
 
 #ifdef HAVE_SYS_TYPES_H
 #include <sys/types.h>
@@ -91,7 +91,7 @@ PyDoc_STRVAR(module_doc,
 "   I/O classes. open() uses the file's blksize (as obtained by os.stat) if\n"
 "   possible.\n"
     );
-
+
 
 /*
  * The main open() function
@@ -509,7 +509,7 @@ _io_text_encoding_impl(PyObject *module, PyObject *encoding, int stacklevel)
 /*[clinic end generated code: output=91b2cfea6934cc0c input=bf70231213e2a7b4]*/
 {
     if (encoding == NULL || encoding == Py_None) {
-        PyInterpreterState *interp = _PyInterpreterState_GET();
+        PyInterpreterState *interp = PyInterpreterState_Get();
         if (_PyInterpreterState_GetConfig(interp)->warn_default_encoding) {
             if (PyErr_WarnEx(PyExc_EncodingWarning,
                              "'encoding' argument not specified", stacklevel)) {
@@ -542,7 +542,7 @@ _io_open_code_impl(PyObject *module, PyObject *path)
 {
     return PyFile_OpenCodeObject(path);
 }
-
+
 /*
  * Private helpers for the io module.
  */

Modules/_io/textio.c

@@ -12,7 +12,7 @@
 #include "pycore_long.h"          // _PyLong_GetZero()
 #include "pycore_fileutils.h"     // _Py_GetLocaleEncoding()
 #include "pycore_object.h"
-#include "pycore_pystate.h"       // _PyInterpreterState_GET()
+#include "pycore_pystate.h"       // PyInterpreterState_Get()
 #include "structmember.h"         // PyMemberDef
 #include "_iomodule.h"
 
@@ -996,7 +996,7 @@ io_check_errors(PyObject *errors)
 {
     assert(errors != NULL && errors != Py_None);
 
-    PyInterpreterState *interp = _PyInterpreterState_GET();
+    PyInterpreterState *interp = PyInterpreterState_Get();
 #ifndef Py_DEBUG
     /* In release mode, only check in development mode (-X dev) */
     if (!_PyInterpreterState_GetConfig(interp)->dev_mode) {
@@ -1086,7 +1086,7 @@ _io_TextIOWrapper___init___impl(textio *self, PyObject *buffer,
     self->detached = 0;
 
     if (encoding == NULL) {
-        PyInterpreterState *interp = _PyInterpreterState_GET();
+        PyInterpreterState *interp = PyInterpreterState_Get();
         if (_PyInterpreterState_GetConfig(interp)->warn_default_encoding) {
             if (PyErr_WarnEx(PyExc_EncodingWarning,
                              "'encoding' argument not specified", 1)) {

Modules/_threadmodule.c

@@ -850,7 +850,7 @@ local_clear(localobject *self)
     Py_CLEAR(self->wr_callback);
     /* Remove all strong references to dummies from the thread states */
     if (self->key) {
-        PyInterpreterState *interp = _PyInterpreterState_GET();
+        PyInterpreterState *interp = PyInterpreterState_Get();
         PyThreadState *tstate = PyInterpreterState_ThreadHead(interp);
         for(; tstate; tstate = PyThreadState_Next(tstate)) {
             if (tstate->dict == NULL) {
@@ -1139,7 +1139,7 @@ thread_PyThread_start_new_thread(PyObject *self, PyObject *fargs)
         return NULL;
     }
 
-    PyInterpreterState *interp = _PyInterpreterState_GET();
+    PyInterpreterState *interp = PyInterpreterState_Get();
     if (interp->config._isolated_interpreter) {
         PyErr_SetString(PyExc_RuntimeError,
                         "thread is not supported for isolated subinterpreters");
@@ -1150,7 +1150,7 @@ thread_PyThread_start_new_thread(PyObject *self, PyObject *fargs)
     if (boot == NULL) {
         return PyErr_NoMemory();
     }
-    boot->interp = _PyInterpreterState_GET();
+    boot->interp = PyInterpreterState_Get();
     boot->tstate = _PyThreadState_Prealloc(boot->interp);
     if (boot->tstate == NULL) {
         PyMem_Free(boot);
@@ -1278,7 +1278,7 @@ particular thread within a system.");
 static PyObject *
 thread__count(PyObject *self, PyObject *Py_UNUSED(ignored))
 {
-    PyInterpreterState *interp = _PyInterpreterState_GET();
+    PyInterpreterState *interp = PyInterpreterState_Get();
     return PyLong_FromLong(interp->num_threads);
 }
 

Modules/atexitmodule.c

@@ -9,15 +9,15 @@
 #include "Python.h"
 #include "pycore_initconfig.h"    // _PyStatus_NO_MEMORY
 #include "pycore_interp.h"        // PyInterpreterState.atexit
-#include "pycore_pystate.h"       // _PyInterpreterState_GET
+#include "pycore_pystate.h"       // PyInterpreterState_Get
 
 /* ===================================================================== */
 /* Callback machinery. */
 
 static inline struct atexit_state*
 get_atexit_state(void)
 {
-    PyInterpreterState *interp = _PyInterpreterState_GET();
+    PyInterpreterState *interp = PyInterpreterState_Get();
     return &interp->atexit;
 }
 

Modules/gcmodule.c

@@ -131,7 +131,7 @@ gc_decref(PyGC_Head *g)
 static GCState *
 get_gc_state(void)
 {
-    PyInterpreterState *interp = _PyInterpreterState_GET();
+    PyInterpreterState *interp = PyInterpreterState_Get();
     return &interp->gc;
 }
 

Modules/main.c

@@ -6,7 +6,7 @@
 #include "pycore_interp.h"        // _PyInterpreterState.sysdict
 #include "pycore_pathconfig.h"    // _PyPathConfig_ComputeSysPath0()
 #include "pycore_pylifecycle.h"   // _Py_PreInitializeFromPyArgv()
-#include "pycore_pystate.h"       // _PyInterpreterState_GET()
+#include "pycore_pystate.h"       // PyInterpreterState_Get()
 
 /* Includes for exit_sigint() */
 #include <stdio.h>                // perror()
@@ -534,8 +534,8 @@ pymain_repl(PyConfig *config, int *exitcode)
 static void
 pymain_run_python(int *exitcode)
 {
+    PyInterpreterState *interp = PyInterpreterState_Get();
     PyObject *main_importer_path = NULL;
-    PyInterpreterState *interp = _PyInterpreterState_GET();
     /* pymain_run_stdin() modify the config */
     PyConfig *config = (PyConfig*)_PyInterpreterState_GetConfig(interp);
 

Modules/posixmodule.c

@@ -27,7 +27,7 @@
 #include "pycore_ceval.h"         // _PyEval_ReInitThreads()
 #include "pycore_import.h"        // _PyImport_ReInitLock()
 #include "pycore_initconfig.h"    // _PyStatus_EXCEPTION()
-#include "pycore_pystate.h"       // _PyInterpreterState_GET()
+#include "pycore_pystate.h"       // PyInterpreterState_Get()
 
 #include "structmember.h"         // PyMemberDef
 #ifndef MS_WINDOWS
@@ -555,7 +555,7 @@ run_at_forkers(PyObject *lst, int reverse)
 void
 PyOS_BeforeFork(void)
 {
-    run_at_forkers(_PyInterpreterState_GET()->before_forkers, 1);
+    run_at_forkers(PyInterpreterState_Get()->before_forkers, 1);
 
     _PyImport_AcquireLock();
 }
@@ -566,7 +566,7 @@ PyOS_AfterFork_Parent(void)
     if (_PyImport_ReleaseLock() <= 0)
         Py_FatalError("failed releasing import lock after fork");
 
-    run_at_forkers(_PyInterpreterState_GET()->after_forkers_parent, 0);
+    run_at_forkers(PyInterpreterState_Get()->after_forkers_parent, 0);
 }
 
 void
@@ -6665,7 +6665,7 @@ os_register_at_fork_impl(PyObject *module, PyObject *before,
         check_null_or_callable(after_in_parent, "after_in_parent")) {
         return NULL;
     }
-    interp = _PyInterpreterState_GET();
+    interp = PyInterpreterState_Get();
 
     if (register_at_forker(&interp->before_forkers, before)) {
         return NULL;
@@ -6695,8 +6695,7 @@ os_fork1_impl(PyObject *module)
 /*[clinic end generated code: output=0de8e67ce2a310bc input=12db02167893926e]*/
 {
     pid_t pid;
-
-    if (_PyInterpreterState_GET() != PyInterpreterState_Main()) {
+    if (PyInterpreterState_Get() != PyInterpreterState_Main()) {
         PyErr_SetString(PyExc_RuntimeError, "fork not supported for subinterpreters");
         return NULL;
     }
@@ -6730,7 +6729,7 @@ os_fork_impl(PyObject *module)
 /*[clinic end generated code: output=3626c81f98985d49 input=13c956413110eeaa]*/
 {
     pid_t pid;
-    PyInterpreterState *interp = _PyInterpreterState_GET();
+    PyInterpreterState *interp = PyInterpreterState_Get();
     if (interp->config._isolated_interpreter) {
         PyErr_SetString(PyExc_RuntimeError,
                         "fork not supported for isolated subinterpreters");
@@ -7348,7 +7347,7 @@ os_forkpty_impl(PyObject *module)
     int master_fd = -1;
     pid_t pid;
 
-    if (_PyInterpreterState_GET() != PyInterpreterState_Main()) {
+    if (PyInterpreterState_Get() != PyInterpreterState_Main()) {
         PyErr_SetString(PyExc_RuntimeError, "fork not supported for subinterpreters");
         return NULL;
     }

Modules/signalmodule.c

@@ -1662,8 +1662,8 @@ signal_module_exec(PyObject *m)
     }
 #endif
 
-    PyThreadState *tstate = _PyThreadState_GET();
-    if (_Py_IsMainInterpreter(tstate->interp)) {
+    PyInterpreterState *interp = PyInterpreterState_Get();
+    if (_Py_IsMainInterpreter(interp)) {
         if (signal_get_set_handlers(state, d) < 0) {
             return -1;
         }
@@ -2039,8 +2039,7 @@ _PySignal_AfterFork(void)
 int
 _PyOS_IsMainThread(void)
 {
-    PyInterpreterState *interp = _PyInterpreterState_GET();
-    return _Py_ThreadCanHandleSignals(interp);
+    return _Py_ThreadCanHandleSignals(PyInterpreterState_Get());
 }
 
 #ifdef MS_WINDOWS

Objects/floatobject.c

@@ -29,11 +29,10 @@ class float "PyObject *" "&PyFloat_Type"
 
 
 #if PyFloat_MAXFREELIST > 0
-static struct _Py_float_state *
+static inline struct _Py_float_state *
 get_float_state(void)
 {
-    PyInterpreterState *interp = _PyInterpreterState_GET();
-    return &interp->float_state;
+    return &_PyInterpreterState_GET()->float_state;
 }
 #endif
 

Objects/frameobject.c

@@ -21,7 +21,7 @@ static PyMemberDef frame_memberlist[] = {
 };
 
 #if PyFrame_MAXFREELIST > 0
-static struct _Py_frame_state *
+static inline struct _Py_frame_state *
 get_frame_state(void)
 {
     PyInterpreterState *interp = _PyInterpreterState_GET();

Objects/unicodeobject.c

@@ -232,7 +232,7 @@ static int unicode_is_singleton(PyObject *unicode);
 #endif
 
 
-static struct _Py_unicode_state*
+static inline struct _Py_unicode_state*
 get_unicode_state(void)
 {
     PyInterpreterState *interp = _PyInterpreterState_GET();

Parser/asdl_c.py

@@ -1484,7 +1484,7 @@ def generate_module_def(mod, metadata, f, internal_h):
         #include "pycore_ast.h"
         #include "pycore_ast_state.h"       // struct ast_state
         #include "pycore_interp.h"          // _PyInterpreterState.ast
-        #include "pycore_pystate.h"         // _PyInterpreterState_GET()
+        #include "pystate.h"                // PyInterpreterState_Get()
         #include "structmember.h"
         #include <stddef.h>
 
@@ -1494,7 +1494,7 @@ def generate_module_def(mod, metadata, f, internal_h):
         static struct ast_state*
         get_ast_state(void)
         {
-            PyInterpreterState *interp = _PyInterpreterState_GET();
+            PyInterpreterState *interp = PyInterpreterState_Get();
             struct ast_state *state = &interp->ast;
             if (!init_types(state)) {
                 return NULL;