Add _PyThreadState_SwapNoGIL().

Include/internal/pycore_ceval.h

@@ -102,6 +102,7 @@ extern void _PyEval_FiniGIL(PyInterpreterState *interp);
 
 extern void _PyEval_AcquireLock(PyThreadState *tstate);
 extern void _PyEval_ReleaseLock(PyThreadState *tstate);
+extern PyThreadState * _PyThreadState_SwapNoGIL(PyThreadState *);
 
 extern void _PyEval_DeactivateOpCache(void);
 

Python/ceval_gil.c

@@ -631,7 +631,7 @@ PyEval_AcquireThread(PyThreadState *tstate)
 
     take_gil(tstate);
 
-    if (_PyThreadState_Swap(tstate->interp->runtime, tstate) != NULL) {
+    if (_PyThreadState_SwapNoGIL(tstate) != NULL) {
         Py_FatalError("non-NULL old thread state");
     }
 }
@@ -641,8 +641,7 @@ PyEval_ReleaseThread(PyThreadState *tstate)
 {
     assert(is_tstate_valid(tstate));
 
-    _PyRuntimeState *runtime = tstate->interp->runtime;
-    PyThreadState *new_tstate = _PyThreadState_Swap(runtime, NULL);
+    PyThreadState *new_tstate = _PyThreadState_SwapNoGIL(NULL);
     if (new_tstate != tstate) {
         Py_FatalError("wrong thread state");
     }
@@ -690,8 +689,7 @@ _PyEval_SignalAsyncExc(PyInterpreterState *interp)
 PyThreadState *
 PyEval_SaveThread(void)
 {
-    _PyRuntimeState *runtime = &_PyRuntime;
-    PyThreadState *tstate = _PyThreadState_Swap(runtime, NULL);
+    PyThreadState *tstate = _PyThreadState_SwapNoGIL(NULL);
     _Py_EnsureTstateNotNULL(tstate);
 
     struct _ceval_state *ceval = &tstate->interp->ceval;
@@ -707,7 +705,7 @@ PyEval_RestoreThread(PyThreadState *tstate)
 
     take_gil(tstate);
 
-    _PyThreadState_Swap(tstate->interp->runtime, tstate);
+    _PyThreadState_SwapNoGIL(tstate);
 }
 
 
@@ -1011,7 +1009,7 @@ _Py_HandlePending(PyThreadState *tstate)
     /* GIL drop request */
     if (_Py_atomic_load_relaxed_int32(&interp_ceval_state->gil_drop_request)) {
         /* Give another thread a chance */
-        if (_PyThreadState_Swap(runtime, NULL) != tstate) {
+        if (_PyThreadState_SwapNoGIL(NULL) != tstate) {
             Py_FatalError("tstate mix-up");
         }
         drop_gil(interp_ceval_state, tstate);
@@ -1020,7 +1018,7 @@ _Py_HandlePending(PyThreadState *tstate)
 
         take_gil(tstate);
 
-        if (_PyThreadState_Swap(runtime, tstate) != NULL) {
+        if (_PyThreadState_SwapNoGIL(tstate) != NULL) {
             Py_FatalError("orphan tstate");
         }
     }

Python/pylifecycle.c

@@ -591,6 +591,7 @@ init_interp_create_gil(PyThreadState *tstate, int own_gil)
 
     /* finalize_interp_delete() comment explains why _PyEval_FiniGIL() is
        only called here. */
+    // XXX This is broken with a per-interpreter GIL.
     _PyEval_FiniGIL(tstate->interp);
 
     /* Auto-thread-state API */
@@ -645,7 +646,8 @@ pycore_create_interpreter(_PyRuntimeState *runtime,
         return _PyStatus_ERR("can't make first thread");
     }
     _PyThreadState_Bind(tstate);
-    (void) PyThreadState_Swap(tstate);
+    // XXX For now we do this before the GIL is created.
+    (void) _PyThreadState_SwapNoGIL(tstate);
 
     status = init_interp_create_gil(tstate, config.own_gil);
     if (_PyStatus_EXCEPTION(status)) {
@@ -2025,11 +2027,14 @@ new_interpreter(PyThreadState **tstate_p, const PyInterpreterConfig *config)
     }
     _PyThreadState_Bind(tstate);
 
-    PyThreadState *save_tstate = PyThreadState_Swap(tstate);
+    // XXX For now we do this before the GIL is created.
+    PyThreadState *save_tstate = _PyThreadState_SwapNoGIL(tstate);
+    int has_gil = 0;
 
     /* Copy the current interpreter config into the new interpreter */
     const PyConfig *src_config;
     if (save_tstate != NULL) {
+        _PyEval_ReleaseLock(save_tstate);
         src_config = _PyInterpreterState_GetConfig(save_tstate->interp);
     }
     else
@@ -2053,6 +2058,7 @@ new_interpreter(PyThreadState **tstate_p, const PyInterpreterConfig *config)
     if (_PyStatus_EXCEPTION(status)) {
         goto error;
     }
+    has_gil = 1;
 
     status = pycore_interp_init(tstate);
     if (_PyStatus_EXCEPTION(status)) {
@@ -2072,7 +2078,12 @@ new_interpreter(PyThreadState **tstate_p, const PyInterpreterConfig *config)
 
     /* Oops, it didn't work.  Undo it all. */
     PyErr_PrintEx(0);
-    PyThreadState_Swap(save_tstate);
+    if (has_gil) {
+        PyThreadState_Swap(save_tstate);
+    }
+    else {
+        _PyThreadState_SwapNoGIL(save_tstate);
+    }
     PyThreadState_Clear(tstate);
     PyThreadState_Delete(tstate);
     PyInterpreterState_Delete(interp);

Python/pystate.c

@@ -1863,17 +1863,11 @@ PyThreadState_Get(void)
 }
 
 
-PyThreadState *
-_PyThreadState_Swap(_PyRuntimeState *runtime, PyThreadState *newts)
+static void
+_swap_thread_states(_PyRuntimeState *runtime,
+                    PyThreadState *oldts, PyThreadState *newts)
 {
-#if defined(Py_DEBUG)
-    /* This can be called from PyEval_RestoreThread(). Similar
-       to it, we need to ensure errno doesn't change.
-    */
-    int err = errno;
-#endif
-    PyThreadState *oldts = current_fast_get(runtime);
-
+    // XXX Do this only if oldts != NULL?
     current_fast_clear(runtime);
 
     if (oldts != NULL) {
@@ -1887,13 +1881,41 @@ _PyThreadState_Swap(_PyRuntimeState *runtime, PyThreadState *newts)
         current_fast_set(runtime, newts);
         tstate_activate(newts);
     }
+}
+
+PyThreadState *
+_PyThreadState_SwapNoGIL(PyThreadState *newts)
+{
+#if defined(Py_DEBUG)
+    /* This can be called from PyEval_RestoreThread(). Similar
+       to it, we need to ensure errno doesn't change.
+    */
+    int err = errno;
+#endif
+
+    PyThreadState *oldts = current_fast_get(&_PyRuntime);
+    _swap_thread_states(&_PyRuntime, oldts, newts);
 
 #if defined(Py_DEBUG)
     errno = err;
 #endif
     return oldts;
 }
 
+PyThreadState *
+_PyThreadState_Swap(_PyRuntimeState *runtime, PyThreadState *newts)
+{
+    PyThreadState *oldts = current_fast_get(runtime);
+    if (oldts != NULL) {
+        _PyEval_ReleaseLock(oldts);
+    }
+    _swap_thread_states(runtime, oldts, newts);
+    if (newts != NULL) {
+        _PyEval_AcquireLock(newts);
+    }
+    return oldts;
+}
+
 PyThreadState *
 PyThreadState_Swap(PyThreadState *newts)
 {