python
diff --git a/‎Include/cpython/pystate.h
Copy file name to clipboardExpand all lines: Include/cpython/pystate.h
+1Lines changed: 1 addition & 0 deletions b/‎Include/cpython/pystate.h
Copy file name to clipboardExpand all lines: Include/cpython/pystate.h
+1Lines changed: 1 addition & 0 deletions
diff --git a/‎Include/internal/pycore_pystate.h
Copy file name to clipboardExpand all lines: Include/internal/pycore_pystate.h
+3-4Lines changed: 3 additions & 4 deletions b/‎Include/internal/pycore_pystate.h
Copy file name to clipboardExpand all lines: Include/internal/pycore_pystate.h
+3-4Lines changed: 3 additions & 4 deletions
diff --git a/‎Python/pystate.c
Copy file name to clipboardExpand all lines: Python/pystate.c
+7-3Lines changed: 7 additions & 3 deletions b/‎Python/pystate.c
Copy file name to clipboardExpand all lines: Python/pystate.c
+7-3Lines changed: 7 additions & 3 deletions
@@ -46,6 +46,7 @@ typedef struct _err_stackitem {
 
 } _PyErr_StackItem;
 
+extern __thread struct _ts *_Py_tls_tstate;
 
 // The PyThreadState typedef is in Include/pystate.h.
 struct _ts {
 
@@ -189,9 +189,8 @@ typedef struct _Py_AuditHookEntry {
 
 struct _gilstate_runtime_state {
     int check_enabled;
-    /* Assuming the current thread holds the GIL, this is the
-       PyThreadState for the current thread. */
-    _Py_atomic_address tstate_current;
+    /* This is the PyThreadState for the thread holding the GIL. */
+    _Py_atomic_address gil_holder;
     /* The single PyInterpreterState used by this process'
        GILState implementation
     */
@@ -337,7 +336,7 @@ _Py_ThreadCanHandlePendingCalls(void)
    The caller must hold the GIL.
 
    See also PyThreadState_Get() and PyThreadState_GET(). */
-#define _PyThreadState_GET() ((PyThreadState*)_Py_atomic_load_relaxed(&_PyRuntime.gilstate.tstate_current))
+#define _PyThreadState_GET() _Py_tls_tstate
 
 /* Redefine PyThreadState_GET() as an alias to _PyThreadState_GET() */
 #undef PyThreadState_GET
 
@@ -32,9 +32,9 @@ extern "C" {
 #endif
 
 #define _PyRuntimeGILState_GetThreadState(gilstate) \
-    ((PyThreadState*)_Py_atomic_load_relaxed(&(gilstate)->tstate_current))
+    ((PyThreadState*)_Py_atomic_load_relaxed(&(gilstate)->gil_holder))
 #define _PyRuntimeGILState_SetThreadState(gilstate, value) \
-    _Py_atomic_store_relaxed(&(gilstate)->tstate_current, \
+    _Py_atomic_store_relaxed(&(gilstate)->gil_holder, \
                              (uintptr_t)(value))
 
 static void
@@ -46,6 +46,7 @@ ensure_tstate_not_null(const char *func, PyThreadState *tstate)
     }
 }
 
+__thread struct _ts *_Py_tls_tstate = NULL;
 
 /* Forward declarations */
 static PyThreadState *_PyGILState_GetThisThreadState(struct _gilstate_runtime_state *gilstate);
@@ -860,6 +861,7 @@ _PyThreadState_DeleteCurrent(PyThreadState *tstate)
     struct _gilstate_runtime_state *gilstate = &tstate->interp->runtime->gilstate;
     tstate_delete_common(tstate, gilstate);
     _PyRuntimeGILState_SetThreadState(gilstate, NULL);
+    _Py_tls_tstate = NULL;
     _PyEval_ReleaseLock(tstate);
     PyMem_RawFree(tstate);
 }
@@ -934,9 +936,11 @@ PyThreadState_Get(void)
 PyThreadState *
 _PyThreadState_Swap(struct _gilstate_runtime_state *gilstate, PyThreadState *newts)
 {
-    PyThreadState *oldts = _PyRuntimeGILState_GetThreadState(gilstate);
+    assert(_PyRuntimeGILState_GetThreadState(gilstate) == _Py_tls_tstate);
+    PyThreadState *oldts = _Py_tls_tstate;
 
     _PyRuntimeGILState_SetThreadState(gilstate, newts);
+    _Py_tls_tstate = newts;
     /* It should not be possible for more than one thread state
        to be used for a thread.  Check this the best we can in debug
        builds.