@@ -0,0 +1,74 @@
+#ifndef Py_INTERNAL_BRC_H
+#define Py_INTERNAL_BRC_H
+
+#include <stdint.h>
+#include "pycore_llist.h"           // struct llist_node
+#include "pycore_lock.h"            // PyMutex
+#include "pycore_object_stack.h"    // _PyObjectStack
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef Py_BUILD_CORE
+#  error "this header requires Py_BUILD_CORE define"
+#endif
+
+#ifdef Py_GIL_DISABLED
+
+// Prime number to avoid correlations with memory addresses.
+#define _Py_BRC_NUM_BUCKETS 257
+
+// Hash table bucket
+struct _brc_bucket {
+    // Mutex protects both the bucket and thread state queues in this bucket.
+    PyMutex mutex;
+
+    // Linked list of _PyThreadStateImpl objects hashed to this bucket.
+    struct llist_node root;
+};
+
+// Per-interpreter biased reference counting state
+struct _brc_state {
+    // Hash table of thread states by thread-id. Thread states within a bucket
+    // are chained using a doubly-linked list.
+    struct _brc_bucket table[_Py_BRC_NUM_BUCKETS];
+};
+
+// Per-thread biased reference counting state
+struct _brc_thread_state {
+    // Linked-list of thread states per hash bucket
+    struct llist_node bucket_node;
+
+    // Thread-id as determined by _PyThread_Id()
+    uintptr_t tid;
+
+    // Objects with refcounts to be merged (protected by bucket mutex)
+    _PyObjectStack objects_to_merge;
+
+    // Local stack of objects to be merged (not accessed by other threads)
+    _PyObjectStack local_objects_to_merge;
+};
+
+// Initialize/finalize the per-thread biased reference counting state
+void _Py_brc_init_thread(PyThreadState *tstate);
+void _Py_brc_remove_thread(PyThreadState *tstate);
+
+// Initialize per-interpreter state
+void _Py_brc_init_state(PyInterpreterState *interp);
+
+void _Py_brc_after_fork(PyInterpreterState *interp);
+
+// Enqueues an object to be merged by it's owning thread (tid). This
+// steals a reference to the object.
+void _Py_brc_queue_object(PyObject *ob);
+
+// Merge the refcounts of queued objects for the current thread.
+void _Py_brc_merge_refcounts(PyThreadState *tstate);
+
+#endif /* Py_GIL_DISABLED */
+
+#ifdef __cplusplus
+}
+#endif
+#endif /* !Py_INTERNAL_BRC_H */
@@ -206,6 +206,7 @@ void _PyEval_FrameClearAndPop(PyThreadState *tstate, _PyInterpreterFrame *frame)
 #define _PY_ASYNC_EXCEPTION_BIT 3
 #define _PY_GC_SCHEDULED_BIT 4
 #define _PY_EVAL_PLEASE_STOP_BIT 5
+#define _PY_EVAL_EXPLICIT_MERGE_BIT 6
 
 /* Reserve a few bits for future use */
 #define _PY_EVAL_EVENTS_BITS 8
 
@@ -201,6 +201,7 @@ struct _is {
 
 #if defined(Py_GIL_DISABLED)
     struct _mimalloc_interp_state mimalloc;
+    struct _brc_state brc;  // biased reference counting state
 #endif
 
     // Per-interpreter state for the obmalloc allocator.  For the main
 
@@ -1,6 +1,8 @@
 #ifndef Py_INTERNAL_OBJECT_STACK_H
 #define Py_INTERNAL_OBJECT_STACK_H
 
+#include "pycore_freelist.h"        // _PyFreeListState
+
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -74,6 +76,10 @@ _PyObjectStack_Pop(_PyObjectStack *stack)
     return obj;
 }
 
+// Merge src into dst, leaving src empty
+extern void
+_PyObjectStack_Merge(_PyObjectStack *dst, _PyObjectStack *src);
+
 // Remove all items from the stack
 extern void
 _PyObjectStack_Clear(_PyObjectStack *stack);
 
@@ -10,6 +10,7 @@ extern "C" {
 
 #include "pycore_freelist.h"      // struct _Py_freelist_state
 #include "pycore_mimalloc.h"      // struct _mimalloc_thread_state
+#include "pycore_brc.h"           // struct _brc_thread_state
 
 
 // Every PyThreadState is actually allocated as a _PyThreadStateImpl. The
@@ -22,6 +23,7 @@ typedef struct _PyThreadStateImpl {
 #ifdef Py_GIL_DISABLED
     struct _mimalloc_thread_state mimalloc;
     struct _Py_freelist_state freelist_state;
+    struct _brc_thread_state brc;
 #endif
 
 } _PyThreadStateImpl;
 
@@ -865,6 +865,7 @@ def __init__(self, f, test):
                     self.test = test
                 def run(self):
                     del self.f
+                    gc_collect()
                     self.test.assertEqual(LAST_FREED, 500)
 
             SetExtra(f.__code__, FREE_INDEX, ctypes.c_voidp(500))
 
@@ -1,8 +1,10 @@
 import threading
 import time
+import unittest
 import weakref
 from concurrent import futures
 from test import support
+from test.support import Py_GIL_DISABLED
 
 
 def mul(x, y):
@@ -83,10 +85,21 @@ def test_no_stale_references(self):
         my_object_collected = threading.Event()
         my_object_callback = weakref.ref(
             my_object, lambda obj: my_object_collected.set())
-        # Deliberately discarding the future.
-        self.executor.submit(my_object.my_method)
+        fut = self.executor.submit(my_object.my_method)
         del my_object
 
+        if Py_GIL_DISABLED:
+            # Due to biased reference counting, my_object might only be
+            # deallocated while the thread that created it runs -- if the
+            # thread is paused waiting on an event, it may not merge the
+            # refcount of the queued object. For that reason, we wait for the
+            # task to finish (so that it's no longer referenced) and force a
+            # GC to ensure that it is collected.
+            fut.result()  # Wait for the task to finish.
+            support.gc_collect()
+        else:
+            del fut  # Deliberately discard the future.
+
         collected = my_object_collected.wait(timeout=support.SHORT_TIMEOUT)
         self.assertTrue(collected,
                         "Stale reference not collected within timeout.")
 
@@ -98,6 +98,7 @@ def test_ressources_gced_in_workers(self):
 
         # explicitly destroy the object to ensure that EventfulGCObj.__del__()
         # is called while manager is still running.
+        support.gc_collect()
         obj = None
         support.gc_collect()
 
 
@@ -405,6 +405,7 @@ PYTHON_OBJS=	\
 		Python/ast_opt.o \
 		Python/ast_unparse.o \
 		Python/bltinmodule.o \
+		Python/brc.o \
 		Python/ceval.o \
 		Python/codecs.o \
 		Python/compile.o \
@@ -1081,6 +1082,7 @@ PYTHON_HEADERS= \
 		$(srcdir)/Include/internal/pycore_atexit.h \
 		$(srcdir)/Include/internal/pycore_bitutils.h \
 		$(srcdir)/Include/internal/pycore_blocks_output_buffer.h \
+		$(srcdir)/Include/internal/pycore_brc.h \
 		$(srcdir)/Include/internal/pycore_bytes_methods.h \
 		$(srcdir)/Include/internal/pycore_bytesobject.h \
 		$(srcdir)/Include/internal/pycore_call.h \
 
@@ -637,6 +637,10 @@ PyOS_AfterFork_Child(void)
     tstate->native_thread_id = PyThread_get_thread_native_id();
 #endif
 
+#ifdef Py_GIL_DISABLED
+    _Py_brc_after_fork(tstate->interp);
+#endif
+
     status = _PyEval_ReInitThreads(tstate);
     if (_PyStatus_EXCEPTION(status)) {
         goto fatal_error;