[3.14] gh-140263: Fix data race in test_lock_two_threads (gh-140264) (gh-140369)

Clang-20 detects a data race between the unlock and the non-atomic
read of the lock state. Use a relaxed load for the assertion to avoid
the race.
(cherry picked from commit f11ec6e643f54f4ee698f7dfc878812a315f2af4)

Co-authored-by: Sam Gross <colesbury@gmail.com>

diff --git a/Modules/_testinternalcapi/test_lock.c b/Modules/_testinternalcapi/test_lock.c
index 8d8cb992b0e07fc66e3282d00e005910827ac630..ded76ca9fe6819436ec608e2211fb4674b0badb8 100644 (file)
--- a/Modules/_testinternalcapi/test_lock.c
+++ b/Modules/_testinternalcapi/test_lock.c
@@ -91,7 +91,8 @@ test_lock_two_threads(PyObject *self, PyObject *obj)
     } while (v != 3 && iters < 200);
 
     // both the "locked" and the "has parked" bits should be set
-    assert(test_data.m._bits == 3);
+    v = _Py_atomic_load_uint8_relaxed(&test_data.m._bits);
+    assert(v == 3);
 
     PyMutex_Unlock(&test_data.m);
     PyEvent_Wait(&test_data.done);