diff --git a/Include/Python.h b/Include/Python.h
index 769ec49779c..7fdb9df7065 100644
--- a/Include/Python.h
+++ b/Include/Python.h
@@ -131,6 +131,7 @@
 #include "pyerrors.h"
 
 #include "cpython/initconfig.h"
+#include "pythread.h"
 #include "pystate.h"
 #include "context.h"
 
diff --git a/Include/pystate.h b/Include/pystate.h
index c46d3fef15d..65b0a24e87f 100644
--- a/Include/pystate.h
+++ b/Include/pystate.h
@@ -7,8 +7,6 @@
 extern "C" {
 #endif
 
-#include "pythread.h"
-
 /* This limitation is for performance and simplicity. If needed it can be
 removed (with effort). */
 #define MAX_CO_EXTRA_USERS 255
diff --git a/Modules/_blake2/blake2b_impl.c b/Modules/_blake2/blake2b_impl.c
index 88c103f56c4..7fb1296f8b2 100644
--- a/Modules/_blake2/blake2b_impl.c
+++ b/Modules/_blake2/blake2b_impl.c
@@ -15,7 +15,6 @@
 
 #include "Python.h"
 #include "pystrhex.h"
-#include "pythread.h"
 
 #include "../hashlib.h"
 #include "blake2ns.h"
diff --git a/Modules/_blake2/blake2s_impl.c b/Modules/_blake2/blake2s_impl.c
index c7c7eafdee2..e3e90d0587b 100644
--- a/Modules/_blake2/blake2s_impl.c
+++ b/Modules/_blake2/blake2s_impl.c
@@ -15,7 +15,6 @@
 
 #include "Python.h"
 #include "pystrhex.h"
-#include "pythread.h"
 
 #include "../hashlib.h"
 #include "blake2ns.h"
diff --git a/Modules/_bz2module.c b/Modules/_bz2module.c
index 008aef0555f..d46f96cc866 100644
--- a/Modules/_bz2module.c
+++ b/Modules/_bz2module.c
@@ -5,8 +5,6 @@
 #include "Python.h"
 #include "structmember.h"
 
-#include "pythread.h"
-
 #include <bzlib.h>
 #include <stdio.h>
 
diff --git a/Modules/_decimal/_decimal.c b/Modules/_decimal/_decimal.c
index cdc942f2158..99f12364ad8 100644
--- a/Modules/_decimal/_decimal.c
+++ b/Modules/_decimal/_decimal.c
@@ -28,7 +28,6 @@
 
 #include <Python.h>
 #include "longintrepr.h"
-#include "pythread.h"
 #include "structmember.h"
 #include "complexobject.h"
 #include "mpdecimal.h"
diff --git a/Modules/_io/bufferedio.c b/Modules/_io/bufferedio.c
index 4ec42eb0108..f243ac84e9e 100644
--- a/Modules/_io/bufferedio.c
+++ b/Modules/_io/bufferedio.c
@@ -11,7 +11,6 @@
 #include "Python.h"
 #include "pycore_object.h"
 #include "structmember.h"
-#include "pythread.h"
 #include "_iomodule.h"
 
 /*[clinic input]
diff --git a/Modules/_lzmamodule.c b/Modules/_lzmamodule.c
index 714458be783..eac36b2d72f 100644
--- a/Modules/_lzmamodule.c
+++ b/Modules/_lzmamodule.c
@@ -9,7 +9,6 @@
 
 #include "Python.h"
 #include "structmember.h"
-#include "pythread.h"
 
 #include <stdarg.h>
 #include <string.h>
diff --git a/Modules/_queuemodule.c b/Modules/_queuemodule.c
index 28bf8991285..062102e5c81 100644
--- a/Modules/_queuemodule.c
+++ b/Modules/_queuemodule.c
@@ -1,6 +1,5 @@
 #include "Python.h"
 #include "structmember.h" /* offsetof */
-#include "pythread.h"
 
 /*[clinic input]
 module _queue
diff --git a/Modules/_sqlite/connection.c b/Modules/_sqlite/connection.c
index 91041b95cd8..87a9f410213 100644
--- a/Modules/_sqlite/connection.c
+++ b/Modules/_sqlite/connection.c
@@ -30,8 +30,6 @@
 #include "prepare_protocol.h"
 #include "util.h"
 
-#include "pythread.h"
-
 #define ACTION_FINALIZE 1
 #define ACTION_RESET 2
 
diff --git a/Modules/_ssl.c b/Modules/_ssl.c
index a471a26e933..d633a06053a 100644
--- a/Modules/_ssl.c
+++ b/Modules/_ssl.c
@@ -18,8 +18,6 @@
 
 #include "Python.h"
 
-#include "pythread.h"
-
 /* Redefined below for Windows debug builds after important #includes */
 #define _PySSL_FIX_ERRNO
 
diff --git a/Modules/_testcapimodule.c b/Modules/_testcapimodule.c
index c97cbe83551..8a76a3b1f24 100644
--- a/Modules/_testcapimodule.c
+++ b/Modules/_testcapimodule.c
@@ -17,7 +17,6 @@
 #include "Python.h"
 #include "datetime.h"
 #include "marshal.h"
-#include "pythread.h"
 #include "structmember.h"
 #include <float.h>
 #include <signal.h>
diff --git a/Modules/_threadmodule.c b/Modules/_threadmodule.c
index 8ff06698a80..6e7ee0f43e3 100644
--- a/Modules/_threadmodule.c
+++ b/Modules/_threadmodule.c
@@ -6,7 +6,6 @@
 #include "pycore_pylifecycle.h"
 #include "pycore_interp.h"       // _PyInterpreterState.num_threads
 #include "pycore_pystate.h"      // _PyThreadState_Init()
-#include "pythread.h"
 #include <stddef.h>              // offsetof()
 
 static PyObject *ThreadError;
diff --git a/Modules/_tkinter.c b/Modules/_tkinter.c
index 199ae4f0db8..f530c5b0eb7 100644
--- a/Modules/_tkinter.c
+++ b/Modules/_tkinter.c
@@ -26,8 +26,6 @@ Copyright (C) 1994 Steen Lumholt.
 #include "Python.h"
 #include <ctype.h>
 
-#include "pythread.h"
-
 #ifdef MS_WINDOWS
 #include <windows.h>
 #endif
diff --git a/Modules/_tracemalloc.c b/Modules/_tracemalloc.c
index 6f3f31a67b5..691de07b9b3 100644
--- a/Modules/_tracemalloc.c
+++ b/Modules/_tracemalloc.c
@@ -4,7 +4,6 @@
 #include "pycore_traceback.h"
 #include "hashtable.h"
 #include "frameobject.h"
-#include "pythread.h"
 #include "osdefs.h"
 
 #include "clinic/_tracemalloc.c.h"
diff --git a/Modules/faulthandler.c b/Modules/faulthandler.c
index ec9debcd948..e7a28503305 100644
--- a/Modules/faulthandler.c
+++ b/Modules/faulthandler.c
@@ -1,7 +1,6 @@
 #include "Python.h"
 #include "pycore_initconfig.h"
 #include "pycore_traceback.h"
-#include "pythread.h"
 #include <signal.h>
 #include <object.h>
 #include <frameobject.h>
diff --git a/Modules/posixmodule.c b/Modules/posixmodule.c
index 692dda314b3..7ac74842152 100644
--- a/Modules/posixmodule.c
+++ b/Modules/posixmodule.c
@@ -38,7 +38,6 @@
 #include "pycore_ceval.h"     /* _PyEval_ReInitThreads() */
 #include "pycore_import.h"    /* _PyImport_ReInitLock() */
 #include "pycore_pystate.h"   /* _PyInterpreterState_GET() */
-#include "pythread.h"
 #include "structmember.h"
 #ifndef MS_WINDOWS
 #  include "posixmodule.h"
diff --git a/Modules/signalmodule.c b/Modules/signalmodule.c
index 69d69ace28e..8348971c353 100644
--- a/Modules/signalmodule.c
+++ b/Modules/signalmodule.c
@@ -104,8 +104,6 @@ class sigset_t_converter(CConverter):
    may not be the thread that received the signal.
 */
 
-#include "pythread.h"
-
 static volatile struct {
     _Py_atomic_int tripped;
     PyObject *func;
diff --git a/Modules/socketmodule.c b/Modules/socketmodule.c
index 722c06e372e..2be0c103f80 100644
--- a/Modules/socketmodule.c
+++ b/Modules/socketmodule.c
@@ -234,13 +234,8 @@ http://cvsweb.netbsd.org/bsdweb.cgi/src/lib/libc/net/getaddrinfo.c.diff?r1=1.82&
 #define RELEASE_GETADDRINFO_LOCK
 #endif
 
-#if defined(USE_GETHOSTBYNAME_LOCK) || defined(USE_GETADDRINFO_LOCK)
-# include "pythread.h"
-#endif
-
-
 #if defined(__APPLE__) || defined(__CYGWIN__) || defined(__NetBSD__)
-# include <sys/ioctl.h>
+#  include <sys/ioctl.h>
 #endif
 
 
diff --git a/Modules/timemodule.c b/Modules/timemodule.c
index e269cd047fb..a0e66ac170b 100644
--- a/Modules/timemodule.c
+++ b/Modules/timemodule.c
@@ -25,13 +25,12 @@
 #endif
 
 #if defined(__WATCOMC__) && !defined(__QNX__)
-#include <i86.h>
+#  include <i86.h>
 #else
-#ifdef MS_WINDOWS
-#define WIN32_LEAN_AND_MEAN
-#include <windows.h>
-#include "pythread.h"
-#endif /* MS_WINDOWS */
+#  ifdef MS_WINDOWS
+#    define WIN32_LEAN_AND_MEAN
+#    include <windows.h>
+#  endif /* MS_WINDOWS */
 #endif /* !__WATCOMC__ || __QNX__ */
 
 #ifdef _Py_MEMORY_SANITIZER
diff --git a/Modules/zlibmodule.c b/Modules/zlibmodule.c
index 49d231a4baa..c2900a1243f 100644
--- a/Modules/zlibmodule.c
+++ b/Modules/zlibmodule.c
@@ -10,7 +10,6 @@
 #include "zlib.h"
 
 
-#include "pythread.h"
 #define ENTER_ZLIB(obj) \
     Py_BEGIN_ALLOW_THREADS; \
     PyThread_acquire_lock((obj)->lock, 1); \
diff --git a/Parser/myreadline.c b/Parser/myreadline.c
index 7da8ea83780..04c2793225c 100644
--- a/Parser/myreadline.c
+++ b/Parser/myreadline.c
@@ -19,7 +19,6 @@
 
 PyThreadState* _PyOS_ReadlineTState = NULL;
 
-#include "pythread.h"
 static PyThread_type_lock _PyOS_ReadlineLock = NULL;
 
 int (*PyOS_InputHook)(void) = NULL;
diff --git a/Programs/_testembed.c b/Programs/_testembed.c
index da3e7861d1e..249f7e2304f 100644
--- a/Programs/_testembed.c
+++ b/Programs/_testembed.c
@@ -9,7 +9,6 @@
 #include "pycore_initconfig.h"    // _PyConfig_InitCompatConfig()
 #include "pycore_runtime.h"       // _PyRuntime
 #include <Python.h>
-#include "pythread.h"
 #include <inttypes.h>
 #include <stdio.h>
 #include <wchar.h>
diff --git a/Python/ceval.c b/Python/ceval.c
index 505f05cadd1..5e543567193 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -240,7 +240,6 @@ UNSIGNAL_ASYNC_EXC(PyInterpreterState *interp)
 #ifdef HAVE_ERRNO_H
 #include <errno.h>
 #endif
-#include "pythread.h"
 #include "ceval_gil.h"
 
 static void
diff --git a/Python/import.c b/Python/import.c
index 49334637d67..2f2e9d17061 100644
--- a/Python/import.c
+++ b/Python/import.c
@@ -150,8 +150,6 @@ _PyImportZip_Init(PyThreadState *tstate)
    in different threads to return with a partially loaded module.
    These calls are serialized by the global interpreter lock. */
 
-#include "pythread.h"
-
 static PyThread_type_lock import_lock = 0;
 static unsigned long import_lock_thread = PYTHREAD_INVALID_THREAD_ID;
 static int import_lock_level = 0;
diff --git a/Python/pylifecycle.c b/Python/pylifecycle.c
index 754e7620a0c..974b425712c 100644
--- a/Python/pylifecycle.c
+++ b/Python/pylifecycle.c
@@ -2350,8 +2350,6 @@ Py_ExitStatusException(PyStatus status)
 
 /* Clean up and exit */
 
-#  include "pythread.h"
-
 /* For the atexit module. */
 void _Py_PyAtExit(void (*func)(PyObject *), PyObject *module)
 {
diff --git a/Python/sysmodule.c b/Python/sysmodule.c
index 741979a09ac..63111d58ab2 100644
--- a/Python/sysmodule.c
+++ b/Python/sysmodule.c
@@ -26,7 +26,6 @@ Data members:
 #include "pycore_pymem.h"
 #include "pycore_pystate.h"      // _PyThreadState_GET()
 #include "pycore_tupleobject.h"
-#include "pythread.h"
 #include "pydtrace.h"
 
 #include "osdefs.h"
diff --git a/Python/thread.c b/Python/thread.c
index 12761031866..a10f5728dc0 100644
--- a/Python/thread.c
+++ b/Python/thread.c
@@ -23,8 +23,6 @@
 
 #include <stdlib.h>
 
-#include "pythread.h"
-
 #ifndef _POSIX_THREADS
 
 /* Check if we're running on HP-UX and _SC_THREADS is defined. If so, then