pystate: keep track of attached vs. detached state

This adds a "status" field to each PyThreadState. The GC status will be
useful for implementing stop-the-world garbage collection.

Include/ceval.h

@@ -16,6 +16,7 @@ PyAPI_FUNC(PyObject *) PyEval_EvalCodeEx(PyObject *co,
                                          PyObject *const *kwds, int kwdc,
                                          PyObject *const *defs, int defc,
                                          PyObject *kwdefs, PyObject *closure);
+/* Interface to random parts in ceval.c */
 
 /* PyEval_CallObjectWithKeywords(), PyEval_CallObject(), PyEval_CallFunction
  * and PyEval_CallMethod are deprecated. Since they are officially part of the

Include/cpython/pystate.h

@@ -114,6 +114,9 @@ struct _ts {
     PyThreadState *next;
     PyInterpreterState *interp;
 
+    /* thread status (attached, detached, gc) */
+    int status;
+
     /* Has been initialized to a safe state.
 
        In order to be effective, this must be set to 0 during or right
@@ -164,6 +167,8 @@ struct _ts {
      */
     unsigned long native_thread_id;
 
+    uintptr_t fast_thread_id; /* Thread id used for object ownership */
+
     int trash_delete_nesting;
     PyObject *trash_delete_later;
 

Include/internal/pycore_ceval.h

@@ -152,6 +152,9 @@ extern PyObject* _Py_MakeCoro(PyFunctionObject *func);
 
 extern int _Py_HandlePending(PyThreadState *tstate);
 
+extern void _PyEval_TakeGIL(PyThreadState *tstate);
+extern void _PyEval_DropGIL(PyThreadState *tstate);
+
 
 
 #ifdef __cplusplus

Include/internal/pycore_pystate.h

@@ -10,6 +10,11 @@ extern "C" {
 
 #include "pycore_runtime.h"   /* PyRuntimeState */
 
+enum _threadstatus {
+    _Py_THREAD_DETACHED = 0,
+    _Py_THREAD_ATTACHED = 1,
+    _Py_THREAD_GC = 2
+};
 
 /* Check if the current thread is the main thread.
    Use _Py_IsMainInterpreter() to check if it's the main interpreter. */

Include/object.h

@@ -500,6 +500,35 @@ PyAPI_FUNC(void) _Py_NegativeRefcount(const char *filename, int lineno,
 
 PyAPI_FUNC(void) _Py_Dealloc(PyObject *);
 
+static inline uintptr_t
+_Py_ThreadId(void)
+{
+    // copied from mimalloc-internal.h
+    uintptr_t tid;
+#if defined(_MSC_VER) && defined(_M_X64)
+    tid = __readgsqword(48);
+#elif defined(_MSC_VER) && defined(_M_IX86)
+    tid = __readfsdword(24);
+#elif defined(_MSC_VER) && defined(_M_ARM64)
+    tid = __getReg(18);
+#elif defined(__i386__)
+    __asm__("movl %%gs:0, %0" : "=r" (tid));  // 32-bit always uses GS
+#elif defined(__MACH__) && defined(__x86_64__)
+    __asm__("movq %%gs:0, %0" : "=r" (tid));  // x86_64 macOSX uses GS
+#elif defined(__x86_64__)
+    __asm__("movq %%fs:0, %0" : "=r" (tid));  // x86_64 Linux, BSD uses FS
+#elif defined(__arm__)
+    __asm__ ("mrc p15, 0, %0, c13, c0, 3\nbic %0, %0, #3" : "=r" (tid));
+#elif defined(__aarch64__) && defined(__APPLE__)
+    __asm__ ("mrs %0, tpidrro_el0" : "=r" (tid));
+#elif defined(__aarch64__)
+    __asm__ ("mrs %0, tpidr_el0" : "=r" (tid));
+#else
+  # error "define _Py_ThreadId for this platform"
+#endif
+  return tid;
+}
+
 /*
 These are provided as conveniences to Python runtime embedders, so that
 they can have object code that is not dependent on Python compilation flags.

Python/ceval_gil.c

@@ -1004,3 +1004,16 @@ _Py_HandlePending(PyThreadState *tstate)
     return 0;
 }
 
+void
+_PyEval_TakeGIL(PyThreadState *tstate)
+{
+    _PyThreadState_SET(tstate);
+    take_gil(tstate);
+}
+
+void
+_PyEval_DropGIL(PyThreadState *tstate)
+{
+    _PyThreadState_SET(NULL);
+    _PyEval_ReleaseLock(tstate);
+}

Python/pystate.c

@@ -13,6 +13,7 @@
 #include "pycore_pystate.h"       // _PyThreadState_GET()
 #include "pycore_runtime_init.h"  // _PyRuntimeState_INIT
 #include "pycore_sysmodule.h"
+#include "pyatomic.h"
 
 /* --------------------------------------------------------------------------
 CAUTION
@@ -240,6 +241,30 @@ _PyRuntimeState_ReInitThreads(_PyRuntimeState *runtime)
 static void _PyGILState_NoteThreadState(
     struct _gilstate_runtime_state *gilstate, PyThreadState* tstate);
 
+int
+_PyThreadState_GetStatus(PyThreadState *tstate)
+{
+    return _Py_atomic_load_int_relaxed(&tstate->status);
+}
+
+static int
+_PyThreadState_Attach(PyThreadState *tstate)
+{
+    if (_Py_atomic_compare_exchange_int(
+            &tstate->status,
+            _Py_THREAD_DETACHED,
+            _Py_THREAD_ATTACHED)) {
+        return 1;
+    }
+    return 0;
+}
+
+static void
+_PyThreadState_Detach(PyThreadState *tstate)
+{
+    _Py_atomic_store_int(&tstate->status, _Py_THREAD_DETACHED);
+}
+
 PyStatus
 _PyInterpreterState_Enable(_PyRuntimeState *runtime)
 {
@@ -517,13 +542,14 @@ PyInterpreterState_Delete(PyInterpreterState *interp)
 {
     _PyRuntimeState *runtime = interp->runtime;
     struct pyinterpreters *interpreters = &runtime->interpreters;
-    zapthreads(interp, 0);
-
-    _PyEval_FiniState(&interp->ceval);
 
     /* Delete current thread. After this, many C API calls become crashy. */
     _PyThreadState_Swap(&runtime->gilstate, NULL);
 
+    zapthreads(interp, 0);
+
+    _PyEval_FiniState(&interp->ceval);
+
     HEAD_LOCK(runtime);
     PyInterpreterState **p;
     for (p = &interpreters->head; ; p = &(*p)->next) {
@@ -910,6 +936,7 @@ _PyThreadState_Init(PyThreadState *tstate)
 void
 _PyThreadState_SetCurrent(PyThreadState *tstate)
 {
+    tstate->fast_thread_id = _Py_ThreadId();
     _PyGILState_NoteThreadState(&tstate->interp->runtime->gilstate, tstate);
 }
 
@@ -1094,15 +1121,25 @@ PyThreadState_Clear(PyThreadState *tstate)
 /* Common code for PyThreadState_Delete() and PyThreadState_DeleteCurrent() */
 static void
 tstate_delete_common(PyThreadState *tstate,
-                     struct _gilstate_runtime_state *gilstate)
+                     struct _gilstate_runtime_state *gilstate,
+                     int is_current)
 {
+    assert(is_current ? tstate->status == _Py_THREAD_ATTACHED
+                      : tstate->status != _Py_THREAD_ATTACHED);
+
     _Py_EnsureTstateNotNULL(tstate);
     PyInterpreterState *interp = tstate->interp;
     if (interp == NULL) {
         Py_FatalError("NULL interpreter");
     }
-    _PyRuntimeState *runtime = interp->runtime;
 
+    if (gilstate->autoInterpreterState &&
+        PyThread_tss_get(&gilstate->autoTSSkey) == tstate)
+    {
+        PyThread_tss_set(&gilstate->autoTSSkey, NULL);
+    }
+
+    _PyRuntimeState *runtime = interp->runtime;
     HEAD_LOCK(runtime);
     if (tstate->prev) {
         tstate->prev->next = tstate->next;
@@ -1115,10 +1152,8 @@ tstate_delete_common(PyThreadState *tstate,
     }
     HEAD_UNLOCK(runtime);
 
-    if (gilstate->autoInterpreterState &&
-        PyThread_tss_get(&gilstate->autoTSSkey) == tstate)
-    {
-        PyThread_tss_set(&gilstate->autoTSSkey, NULL);
+    if (is_current) {
+        _PyThreadState_SET(NULL);
     }
     _PyStackChunk *chunk = tstate->datastack_chunk;
     tstate->datastack_chunk = NULL;
@@ -1138,7 +1173,7 @@ _PyThreadState_Delete(PyThreadState *tstate, int check_current)
             _Py_FatalErrorFormat(__func__, "tstate %p is still current", tstate);
         }
     }
-    tstate_delete_common(tstate, gilstate);
+    tstate_delete_common(tstate, gilstate, 0);
     free_threadstate(tstate);
 }
 
@@ -1155,7 +1190,7 @@ _PyThreadState_DeleteCurrent(PyThreadState *tstate)
 {
     _Py_EnsureTstateNotNULL(tstate);
     struct _gilstate_runtime_state *gilstate = &tstate->interp->runtime->gilstate;
-    tstate_delete_common(tstate, gilstate);
+    tstate_delete_common(tstate, gilstate, 1);
     _PyRuntimeGILState_SetThreadState(gilstate, NULL);
     _PyEval_ReleaseLock(tstate);
     free_threadstate(tstate);
@@ -1230,9 +1265,36 @@ PyThreadState_Get(void)
 PyThreadState *
 _PyThreadState_Swap(struct _gilstate_runtime_state *gilstate, PyThreadState *newts)
 {
-    PyThreadState *oldts = _PyRuntimeGILState_GetThreadState(gilstate);
+    PyThreadState *oldts = _Py_current_tstate;
+
+#if defined(Py_DEBUG)
+    // The new thread-state should correspond to the current native thread
+    // XXX: breaks subinterpreter tests
+    if (newts && newts->fast_thread_id != _Py_ThreadId()) {
+        Py_FatalError("Invalid thread state for this thread");
+    }
+#endif
+
+    if (oldts != NULL) {
+        int status = _Py_atomic_load_int(&oldts->status);
+        assert(status == _Py_THREAD_ATTACHED || status == _Py_THREAD_GC);
+
+        if (status == _Py_THREAD_ATTACHED) {
+            _PyThreadState_Detach(oldts);
+        }
+    }
+
+    _Py_current_tstate = newts;
+
+    if (newts) {
+        int attached = _PyThreadState_Attach(newts);
+        if (!attached) {
+            // _PyThreadState_GC_Park(newts);
+        }
+
+        assert(_Py_atomic_load_int(&newts->status) == _Py_THREAD_ATTACHED);
+    }
 
-    _PyRuntimeGILState_SetThreadState(gilstate, 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.
@@ -1243,8 +1305,7 @@ _PyThreadState_Swap(struct _gilstate_runtime_state *gilstate, PyThreadState *new
            to it, we need to ensure errno doesn't change.
         */
         int err = errno;
-        PyThreadState *check = _PyGILState_GetThisThreadState(gilstate);
-        if (check && check->interp == newts->interp && check != newts)
+        if (oldts && oldts->interp == newts->interp && oldts != newts)
             Py_FatalError("Invalid thread state for this thread");
         errno = err;
     }