gh-115999: Enable specialization of CALL instructions in free-threaded builds (#127123)

The CALL family of instructions were mostly thread-safe already and only required a small number of changes, which are documented below.

A few changes were needed to make CALL_ALLOC_AND_ENTER_INIT thread-safe:

Added _PyType_LookupRefAndVersion, which returns the type version corresponding to the returned ref.

Added _PyType_CacheInitForSpecialization, which takes an init method and the corresponding type version and only populates the specialization cache if the current type version matches the supplied version. This prevents potentially caching a stale value in free-threaded builds if we race with an update to __init__.

Only cache __init__ functions that are deferred in free-threaded builds. This ensures that the reference to __init__ that is stored in the specialization cache is valid if the type version guard in _CHECK_AND_ALLOCATE_OBJECT passes.
Fix a bug in _CREATE_INIT_FRAME where the frame is pushed to the stack on failure.

A few other miscellaneous changes were also needed:

Use {LOCK,UNLOCK}_OBJECT in LIST_APPEND. This ensures that the list's per-object lock is held while we are appending to it.

Add missing co_tlbc for _Py_InitCleanup.

Stop/start the world around setting the eval frame hook. This allows us to read interp->eval_frame non-atomically and preserves the behavior of _CHECK_PEP_523 documented below.
This commit is contained in:
mpage 2024-12-03 11:20:20 -08:00 committed by GitHub
parent fc5a0dc224
commit dabcecfd6d
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
11 changed files with 220 additions and 92 deletions

View file

@ -4500,13 +4500,13 @@
JUMP_TO_JUMP_TARGET();
}
PyTypeObject *tp = (PyTypeObject *)callable_o;
if (tp->tp_version_tag != type_version) {
if (FT_ATOMIC_LOAD_UINT32_RELAXED(tp->tp_version_tag) != type_version) {
UOP_STAT_INC(uopcode, miss);
JUMP_TO_JUMP_TARGET();
}
assert(tp->tp_flags & Py_TPFLAGS_INLINE_VALUES);
PyHeapTypeObject *cls = (PyHeapTypeObject *)callable_o;
PyFunctionObject *init_func = (PyFunctionObject *)cls->_spec_cache.init;
PyFunctionObject *init_func = (PyFunctionObject *)FT_ATOMIC_LOAD_PTR_ACQUIRE(cls->_spec_cache.init);
PyCodeObject *code = (PyCodeObject *)init_func->func_code;
if (!_PyThreadState_HasStackSpace(tstate, code->co_framesize + _Py_InitCleanup.co_framesize)) {
UOP_STAT_INC(uopcode, miss);
@ -4537,25 +4537,29 @@
_PyInterpreterFrame *shim = _PyFrame_PushTrampolineUnchecked(
tstate, (PyCodeObject *)&_Py_InitCleanup, 1, frame);
assert(_PyFrame_GetBytecode(shim)[0].op.code == EXIT_INIT_CHECK);
assert(_PyFrame_GetBytecode(shim)[1].op.code == RETURN_VALUE);
stack_pointer = _PyFrame_GetStackPointer(frame);
/* Push self onto stack of shim */
shim->localsplus[0] = PyStackRef_DUP(self[0]);
_PyFrame_SetStackPointer(frame, stack_pointer);
init_frame = _PyEvalFramePushAndInit(
_PyInterpreterFrame *temp = _PyEvalFramePushAndInit(
tstate, init[0], NULL, args-1, oparg+1, NULL, shim);
stack_pointer = _PyFrame_GetStackPointer(frame);
stack_pointer[-2 - oparg].bits = (uintptr_t)init_frame;
stack_pointer += -1 - oparg;
stack_pointer += -2 - oparg;
assert(WITHIN_STACK_BOUNDS());
if (init_frame == NULL) {
if (temp == NULL) {
_PyEval_FrameClearAndPop(tstate, shim);
JUMP_TO_ERROR();
}
init_frame = temp;
frame->return_offset = 1 + INLINE_CACHE_ENTRIES_CALL;
/* Account for pushing the extra frame.
* We don't check recursion depth here,
* as it will be checked after start_frame */
tstate->py_recursion_remaining--;
stack_pointer[0].bits = (uintptr_t)init_frame;
stack_pointer += 1;
assert(WITHIN_STACK_BOUNDS());
break;
}
@ -4908,8 +4912,13 @@
UOP_STAT_INC(uopcode, miss);
JUMP_TO_JUMP_TARGET();
}
if (!LOCK_OBJECT(self_o)) {
UOP_STAT_INC(uopcode, miss);
JUMP_TO_JUMP_TARGET();
}
STAT_INC(CALL, hit);
int err = _PyList_AppendTakeRef((PyListObject *)self_o, PyStackRef_AsPyObjectSteal(arg));
UNLOCK_OBJECT(self_o);
PyStackRef_CLOSE(self);
PyStackRef_CLOSE(callable);
if (err) JUMP_TO_ERROR();