gh-116818: Make sys.settrace, sys.setprofile, and monitoring thre…

…ad-safe (#116775) Makes sys.settrace, sys.setprofile, and monitoring generally thread-safe. Mostly uses a stop-the-world approach and synchronization around the code object's _co_instrumentation_version. There may be a little bit of extra synchronization around the monitoring data that's required to be TSAN clean.
python · Apr 19, 2024 · 07525c9 · 07525c9
1 parent b45af00
commit 07525c9
Show file tree

Hide file tree

Showing 18 changed files with 528 additions and 61 deletions.
diff --git a/Include/cpython/pyatomic.h b/Include/cpython/pyatomic.h
@@ -465,10 +465,16 @@ _Py_atomic_store_ullong_relaxed(unsigned long long *obj,
 static inline void *
 _Py_atomic_load_ptr_acquire(const void *obj);
 
+static inline uintptr_t
+_Py_atomic_load_uintptr_acquire(const uintptr_t *obj);
+
 // Stores `*obj = value` (release operation)
 static inline void
 _Py_atomic_store_ptr_release(void *obj, void *value);
 
+static inline void
+_Py_atomic_store_uintptr_release(uintptr_t *obj, uintptr_t value);
+
 static inline void
 _Py_atomic_store_ssize_release(Py_ssize_t *obj, Py_ssize_t value);
 
@@ -491,6 +497,8 @@ static inline Py_ssize_t
 _Py_atomic_load_ssize_acquire(const Py_ssize_t *obj);
 
 
+
+
 // --- _Py_atomic_fence ------------------------------------------------------
 
 // Sequential consistency fence. C11 fences have complex semantics. When

diff --git a/Include/cpython/pyatomic_gcc.h b/Include/cpython/pyatomic_gcc.h
@@ -492,10 +492,18 @@ static inline void *
 _Py_atomic_load_ptr_acquire(const void *obj)
 { return (void *)__atomic_load_n((void **)obj, __ATOMIC_ACQUIRE); }
 
+static inline uintptr_t
+_Py_atomic_load_uintptr_acquire(const uintptr_t *obj)
+{ return (uintptr_t)__atomic_load_n((uintptr_t *)obj, __ATOMIC_ACQUIRE); }
+
 static inline void
 _Py_atomic_store_ptr_release(void *obj, void *value)
 { __atomic_store_n((void **)obj, value, __ATOMIC_RELEASE); }
 
+static inline void
+_Py_atomic_store_uintptr_release(uintptr_t *obj, uintptr_t value)
+{ __atomic_store_n(obj, value, __ATOMIC_RELEASE); }
+
 static inline void
 _Py_atomic_store_int_release(int *obj, int value)
 { __atomic_store_n(obj, value, __ATOMIC_RELEASE); }

diff --git a/Include/cpython/pyatomic_msc.h b/Include/cpython/pyatomic_msc.h
@@ -914,6 +914,18 @@ _Py_atomic_load_ptr_acquire(const void *obj)
 #endif
 }
 
+static inline uintptr_t
+_Py_atomic_load_uintptr_acquire(const uintptr_t *obj)
+{
+#if defined(_M_X64) || defined(_M_IX86)
+    return *(uintptr_t volatile *)obj;
+#elif defined(_M_ARM64)
+    return (uintptr_t)__ldar64((unsigned __int64 volatile *)obj);
+#else
+#  error "no implementation of _Py_atomic_load_uintptr_acquire"
+#endif
+}
+
 static inline void
 _Py_atomic_store_ptr_release(void *obj, void *value)
 {
@@ -926,6 +938,19 @@ _Py_atomic_store_ptr_release(void *obj, void *value)
 #endif
 }
 
+static inline void
+_Py_atomic_store_uintptr_release(uintptr_t *obj, uintptr_t value)
+{
+#if defined(_M_X64) || defined(_M_IX86)
+    *(uintptr_t volatile *)obj = value;
+#elif defined(_M_ARM64)
+    _Py_atomic_ASSERT_ARG_TYPE(unsigned __int64);
+    __stlr64((unsigned __int64 volatile *)obj, (unsigned __int64)value);
+#else
+#  error "no implementation of _Py_atomic_store_uintptr_release"
+#endif
+}
+
 static inline void
 _Py_atomic_store_int_release(int *obj, int value)
 {

diff --git a/Include/cpython/pyatomic_std.h b/Include/cpython/pyatomic_std.h
@@ -863,6 +863,14 @@ _Py_atomic_load_ptr_acquire(const void *obj)
                                 memory_order_acquire);
 }
 
+static inline uintptr_t
+_Py_atomic_load_uintptr_acquire(const uintptr_t *obj)
+{
+    _Py_USING_STD;
+    return atomic_load_explicit((const _Atomic(uintptr_t)*)obj,
+                                memory_order_acquire);
+}
+
 static inline void
 _Py_atomic_store_ptr_release(void *obj, void *value)
 {
@@ -871,6 +879,14 @@ _Py_atomic_store_ptr_release(void *obj, void *value)
                           memory_order_release);
 }
 
+static inline void
+_Py_atomic_store_uintptr_release(uintptr_t *obj, uintptr_t value)
+{
+    _Py_USING_STD;
+    atomic_store_explicit((_Atomic(uintptr_t)*)obj, value,
+                          memory_order_release);
+}
+
 static inline void
 _Py_atomic_store_int_release(int *obj, int value)
 {

diff --git a/Include/internal/pycore_ceval_state.h b/Include/internal/pycore_ceval_state.h
@@ -63,6 +63,7 @@ struct _ceval_runtime_state {
     } perf;
     /* Pending calls to be made only on the main thread. */
     struct _pending_calls pending_mainthread;
+    PyMutex sys_trace_profile_mutex;
 };
 
 #ifdef PY_HAVE_PERF_TRAMPOLINE

diff --git a/Include/internal/pycore_gc.h b/Include/internal/pycore_gc.h
@@ -93,7 +93,7 @@ static inline void _PyObject_GC_SET_SHARED(PyObject *op) {
  * threads and needs special purpose when freeing due to
  * the possibility of in-flight lock-free reads occurring.
  * Objects with this bit that are GC objects will automatically
- * delay-freed by PyObject_GC_Del.  */
+ * delay-freed by PyObject_GC_Del. */
 static inline int _PyObject_GC_IS_SHARED_INLINE(PyObject *op) {
     return (op->ob_gc_bits & _PyGC_BITS_SHARED_INLINE) != 0;
 }

diff --git a/Include/internal/pycore_pyatomic_ft_wrappers.h b/Include/internal/pycore_pyatomic_ft_wrappers.h
@@ -26,20 +26,34 @@ extern "C" {
     _Py_atomic_load_ssize_relaxed(&value)
 #define FT_ATOMIC_STORE_PTR(value, new_value) \
     _Py_atomic_store_ptr(&value, new_value)
+#define FT_ATOMIC_LOAD_PTR_ACQUIRE(value) \
+    _Py_atomic_load_ptr_acquire(&value)
+#define FT_ATOMIC_LOAD_UINTPTR_ACQUIRE(value) \
+    _Py_atomic_load_uintptr_acquire(&value)
 #define FT_ATOMIC_STORE_PTR_RELAXED(value, new_value) \
     _Py_atomic_store_ptr_relaxed(&value, new_value)
 #define FT_ATOMIC_STORE_PTR_RELEASE(value, new_value) \
     _Py_atomic_store_ptr_release(&value, new_value)
+#define FT_ATOMIC_STORE_UINTPTR_RELEASE(value, new_value) \
+    _Py_atomic_store_uintptr_release(&value, new_value)
 #define FT_ATOMIC_STORE_SSIZE_RELAXED(value, new_value) \
     _Py_atomic_store_ssize_relaxed(&value, new_value)
+#define FT_ATOMIC_STORE_UINT8_RELAXED(value, new_value) \
+    _Py_atomic_store_uint8_relaxed(&value, new_value)
+
 #else
 #define FT_ATOMIC_LOAD_PTR(value) value
 #define FT_ATOMIC_LOAD_SSIZE(value) value
 #define FT_ATOMIC_LOAD_SSIZE_RELAXED(value) value
 #define FT_ATOMIC_STORE_PTR(value, new_value) value = new_value
+#define FT_ATOMIC_LOAD_PTR_ACQUIRE(value) value
+#define FT_ATOMIC_LOAD_UINTPTR_ACQUIRE(value) value
 #define FT_ATOMIC_STORE_PTR_RELAXED(value, new_value) value = new_value
 #define FT_ATOMIC_STORE_PTR_RELEASE(value, new_value) value = new_value
+#define FT_ATOMIC_STORE_UINTPTR_RELEASE(value, new_value) value = new_value
 #define FT_ATOMIC_STORE_SSIZE_RELAXED(value, new_value) value = new_value
+#define FT_ATOMIC_STORE_UINT8_RELAXED(value, new_value) value = new_value
+
 #endif
 
 #ifdef __cplusplus

diff --git a/Lib/test/test_free_threading/__init__.py b/Lib/test/test_free_threading/__init__.py
@@ -0,0 +1,7 @@
+import os
+
+from test import support
+
+
+def load_tests(*args):
+    return support.load_package_tests(os.path.dirname(__file__), *args)