gh-149570: avoid tier-2 self-link in _COLD_EXIT

Include/internal/pycore_ceval.h

@@ -351,9 +351,10 @@ _PyEval_SpecialMethodCanSuggest(PyObject *self, int oparg);
 #define _PY_EVAL_PLEASE_STOP_BIT (1U << 5)
 #define _PY_EVAL_EXPLICIT_MERGE_BIT (1U << 6)
 #define _PY_EVAL_JIT_INVALIDATE_COLD_BIT (1U << 7)
+#define _PY_EVAL_JIT_FORCE_TIER1_BIT (1U << 8)
 
 /* Reserve a few bits for future use */
-#define _PY_EVAL_EVENTS_BITS 8
+#define _PY_EVAL_EVENTS_BITS 9
 #define _PY_EVAL_EVENTS_MASK ((1 << _PY_EVAL_EVENTS_BITS)-1)
 
 static inline void

Lib/test/test_capi/test_opt.py

@@ -1,5 +1,6 @@
 import contextlib
 import itertools
+import subprocess
 import sys
 import textwrap
 import unittest
@@ -9,7 +10,7 @@
 
 import _opcode
 
-from test.support import (script_helper, requires_specialization,
+from test.support import (SHORT_TIMEOUT, script_helper, requires_specialization,
                           import_helper, Py_GIL_DISABLED, requires_jit_enabled,
                           reset_code)
 
@@ -5994,6 +5995,51 @@ def __init__(self, name):
              PYTHON_JIT_SIDE_EXIT_INITIAL_VALUE="1")
         self.assertEqual(result[0].rc, 0, result)
 
+    def test_side_exit_to_executor_makes_progress(self):
+        script = textwrap.dedent("""
+        classes = []
+
+        def make_iter():
+            class It:
+                def __init__(self):
+                    self.i = 0
+
+                def __iter__(self):
+                    return self
+
+                def __next__(self):
+                    self.i += 1
+                    if self.i > 1000:
+                        raise StopIteration
+                    return self.i
+
+            classes.append(It)
+            return It()
+
+        def f(n):
+            for outer in range(n):
+                for x in make_iter():
+                    pass
+
+        f(200)
+        """)
+        env = os.environ.copy()
+        env.update({
+            "PYTHON_JIT": "1",
+            "PYTHON_JIT_SIDE_EXIT_INITIAL_VALUE": "1",
+        })
+        try:
+            result = subprocess.run(
+                [sys.executable, "-X", "faulthandler", "-c", script],
+                stdout=subprocess.PIPE,
+                stderr=subprocess.PIPE,
+                env=env,
+                timeout=SHORT_TIMEOUT,
+            )
+        except subprocess.TimeoutExpired as exc:
+            self.fail(f"subprocess timed out: {exc}")
+        self.assertEqual(result.returncode, 0, result)
+
     def test_for_iter_gen_cleared_frame_does_not_crash(self):
         # See: https://github.com/python/cpython/issues/145197
         result = script_helper.run_python_until_end('-c', textwrap.dedent("""

Python/bytecodes.c

@@ -6297,6 +6297,16 @@ dummy_func(
             if (target->op.code == ENTER_EXECUTOR) {
                 PyCodeObject *code = _PyFrame_GetCode(frame);
                 executor = code->co_executors->executors[target->op.arg];
+                _PyExecutorObject *previous_executor = _PyExecutor_FromExit(exit);
+                if (!exit->is_control_flow) {
+                    int chain_depth = previous_executor->vm_data.chain_depth + 1;
+                    if (executor == previous_executor ||
+                        chain_depth % MAX_CHAIN_DEPTH == 0) {
+                        /* Avoid tier-2 self-link/wrap that bypasses the eval breaker. */
+                        _Py_set_eval_breaker_bit(tstate, _PY_EVAL_JIT_FORCE_TIER1_BIT);
+                        GOTO_TIER_ONE(target);
+                    }
+                }
                 Py_INCREF(executor);
                 assert(tstate->jit_exit == exit);
                 exit->executor = executor;

Python/ceval_gil.c

@@ -1403,6 +1403,10 @@ _Py_HandlePending(PyThreadState *tstate)
     }
 
 #ifdef _Py_TIER2
+    if ((breaker & _PY_EVAL_JIT_FORCE_TIER1_BIT) != 0) {
+        _Py_unset_eval_breaker_bit(tstate, _PY_EVAL_JIT_FORCE_TIER1_BIT);
+    }
+
     if ((breaker & _PY_EVAL_JIT_INVALIDATE_COLD_BIT) != 0) {
         _Py_unset_eval_breaker_bit(tstate, _PY_EVAL_JIT_INVALIDATE_COLD_BIT);
         _Py_Executors_InvalidateCold(tstate->interp);