gh-101659: Isolate "obmalloc" State to Each Interpreter (gh-101660)

This is strictly about moving the "obmalloc" runtime state from
`_PyRuntimeState` to `PyInterpreterState`.  Doing so improves isolation
between interpreters, specifically most of the memory (incl. objects)
allocated for each interpreter's use.  This is important for a
per-interpreter GIL, but such isolation is valuable even without it.

FWIW, a per-interpreter obmalloc is the proverbial
canary-in-the-coalmine when it comes to the isolation of objects between
interpreters.  Any object that leaks (unintentionally) to another
interpreter is highly likely to cause a crash (on debug builds at
least).  That's a useful thing to know, relative to interpreter
isolation.

Include/cpython/initconfig.h

@@ -245,6 +245,8 @@ PyAPI_FUNC(PyStatus) PyConfig_SetWideStringList(PyConfig *config,
 /* --- PyInterpreterConfig ------------------------------------ */
 
 typedef struct {
+    // XXX "allow_object_sharing"?  "own_objects"?
+    int use_main_obmalloc;
     int allow_fork;
     int allow_exec;
     int allow_threads;
@@ -254,6 +256,7 @@ typedef struct {
 
 #define _PyInterpreterConfig_INIT \
     { \
+        .use_main_obmalloc = 0, \
         .allow_fork = 0, \
         .allow_exec = 0, \
         .allow_threads = 1, \
@@ -263,6 +266,7 @@ typedef struct {
 
 #define _PyInterpreterConfig_LEGACY_INIT \
     { \
+        .use_main_obmalloc = 1, \
         .allow_fork = 1, \
         .allow_exec = 1, \
         .allow_threads = 1, \

Include/cpython/pystate.h

@@ -11,6 +11,10 @@ is available in a given context.  For example, forking the process
 might not be allowed in the current interpreter (i.e. os.fork() would fail).
 */
 
+/* Set if the interpreter share obmalloc runtime state
+   with the main interpreter. */
+#define Py_RTFLAGS_USE_MAIN_OBMALLOC (1UL << 5)
+
 /* Set if import should check a module for subinterpreter support. */
 #define Py_RTFLAGS_MULTI_INTERP_EXTENSIONS (1UL << 8)
 

Include/internal/pycore_interp.h

@@ -23,11 +23,12 @@ extern "C" {
 #include "pycore_function.h"      // FUNC_MAX_WATCHERS
 #include "pycore_genobject.h"     // struct _Py_async_gen_state
 #include "pycore_gc.h"            // struct _gc_runtime_state
+#include "pycore_global_objects.h"  // struct _Py_interp_static_objects
 #include "pycore_import.h"        // struct _import_state
 #include "pycore_instruments.h"   // PY_MONITORING_EVENTS
 #include "pycore_list.h"          // struct _Py_list_state
-#include "pycore_global_objects.h"  // struct _Py_interp_static_objects
 #include "pycore_object_state.h"   // struct _py_object_state
+#include "pycore_obmalloc.h"      // struct obmalloc_state
 #include "pycore_tuple.h"         // struct _Py_tuple_state
 #include "pycore_typeobject.h"    // struct type_cache
 #include "pycore_unicodeobject.h" // struct _Py_unicode_state
@@ -82,6 +83,8 @@ struct _is {
     int _initialized;
     int finalizing;
 
+    struct _obmalloc_state obmalloc;
+
     struct _ceval_state ceval;
     struct _gc_runtime_state gc;
 

Include/internal/pycore_obmalloc.h

@@ -657,8 +657,12 @@ struct _obmalloc_usage {
 #endif /* WITH_PYMALLOC_RADIX_TREE */
 
 
-struct _obmalloc_state {
+struct _obmalloc_global_state {
     int dump_debug_stats;
+    Py_ssize_t interpreter_leaks;
+};
+
+struct _obmalloc_state {
     struct _obmalloc_pools pools;
     struct _obmalloc_mgmt mgmt;
     struct _obmalloc_usage usage;
@@ -675,7 +679,11 @@ void _PyObject_VirtualFree(void *, size_t size);
 
 
 /* This function returns the number of allocated memory blocks, regardless of size */
-PyAPI_FUNC(Py_ssize_t) _Py_GetAllocatedBlocks(void);
+extern Py_ssize_t _Py_GetGlobalAllocatedBlocks(void);
+#define _Py_GetAllocatedBlocks() \
+    _Py_GetGlobalAllocatedBlocks()
+extern Py_ssize_t _PyInterpreterState_GetAllocatedBlocks(PyInterpreterState *);
+extern void _PyInterpreterState_FinalizeAllocatedBlocks(PyInterpreterState *);
 
 
 #ifdef WITH_PYMALLOC

Include/internal/pycore_obmalloc_init.h

@@ -54,9 +54,13 @@ extern "C" {
 #  error "NB_SMALL_SIZE_CLASSES should be less than 64"
 #endif
 
-#define _obmalloc_state_INIT(obmalloc) \
+#define _obmalloc_global_state_INIT \
     { \
         .dump_debug_stats = -1, \
+    }
+
+#define _obmalloc_state_INIT(obmalloc) \
+    { \
         .pools = { \
             .used = _obmalloc_pools_INIT(obmalloc.pools), \
         }, \

Include/internal/pycore_pylifecycle.h

@@ -64,6 +64,7 @@ extern void _PyAtExit_Fini(PyInterpreterState *interp);
 extern void _PyThread_FiniType(PyInterpreterState *interp);
 extern void _Py_Deepfreeze_Fini(void);
 extern void _PyArg_Fini(void);
+extern void _Py_FinalizeAllocatedBlocks(_PyRuntimeState *);
 
 extern PyStatus _PyGILState_Init(PyInterpreterState *interp);
 extern PyStatus _PyGILState_SetTstate(PyThreadState *tstate);

Include/internal/pycore_pystate.h

@@ -33,6 +33,13 @@ _Py_IsMainInterpreter(PyInterpreterState *interp)
     return (interp == _PyInterpreterState_Main());
 }
 
+static inline int
+_Py_IsMainInterpreterFinalizing(PyInterpreterState *interp)
+{
+    return (_PyRuntimeState_GetFinalizing(interp->runtime) != NULL &&
+            interp == &interp->runtime->_main_interpreter);
+}
+
 
 static inline const PyConfig *
 _Py_GetMainConfig(void)

Include/internal/pycore_runtime.h

@@ -21,7 +21,6 @@ extern "C" {
 #include "pycore_pymem.h"           // struct _pymem_allocators
 #include "pycore_pyhash.h"          // struct pyhash_runtime_state
 #include "pycore_pythread.h"        // struct _pythread_runtime_state
-#include "pycore_obmalloc.h"        // struct obmalloc_state
 #include "pycore_signal.h"          // struct _signals_runtime_state
 #include "pycore_time.h"            // struct _time_runtime_state
 #include "pycore_tracemalloc.h"     // struct _tracemalloc_runtime_state
@@ -88,7 +87,7 @@ typedef struct pyruntimestate {
     _Py_atomic_address _finalizing;
 
     struct _pymem_allocators allocators;
-    struct _obmalloc_state obmalloc;
+    struct _obmalloc_global_state obmalloc;
     struct pyhash_runtime_state pyhash_state;
     struct _time_runtime_state time;
     struct _pythread_runtime_state threads;

Include/internal/pycore_runtime_init.h

@@ -29,7 +29,7 @@ extern PyTypeObject _PyExc_MemoryError;
             _pymem_allocators_debug_INIT, \
             _pymem_allocators_obj_arena_INIT, \
         }, \
-        .obmalloc = _obmalloc_state_INIT(runtime.obmalloc), \
+        .obmalloc = _obmalloc_global_state_INIT, \
         .pyhash_state = pyhash_state_INIT, \
         .signals = _signals_RUNTIME_INIT, \
         .interpreters = { \
@@ -93,6 +93,7 @@ extern PyTypeObject _PyExc_MemoryError;
     { \
         .id_refcount = -1, \
         .imports = IMPORTS_INIT, \
+        .obmalloc = _obmalloc_state_INIT(INTERP.obmalloc), \
         .ceval = { \
             .recursion_limit = Py_DEFAULT_RECURSION_LIMIT, \
         }, \