office-gobmx/external/python3/tsan.patch.0
Stephan Bergmann 4a804f87eb external/python3: Avoid "ThreadSanitizer: destroy of a locked mutex"
..when building ExternalProject_python3 with Clang -fsanitize=thread,

> WARNING: ThreadSanitizer: destroy of a locked mutex (pid=973799)
>  #0 in AnnotateRWLockDestroy at /home/sbergman/github.com/llvm/llvm-project/compiler-rt/lib/tsan/rtl/tsan_interface_ann.cpp:184:3 (workdir/UnpackedTarball/python3/python +0x498460)
>  #1 in recreate_gil at workdir/UnpackedTarball/python3/Python/ceval_gil.h:138:5 (workdir/UnpackedTarball/python3/libpython3.8d.so.1.0 +0xe84aa9)
>  #2 in _PyEval_ReInitThreads at workdir/UnpackedTarball/python3/Python/ceval.c:350:5 (workdir/UnpackedTarball/python3/libpython3.8d.so.1.0 +0xe847c7)
>  #3 in PyOS_AfterFork_Child at workdir/UnpackedTarball/python3/./Modules/posixmodule.c:469:5 (workdir/UnpackedTarball/python3/libpython3.8d.so.1.0 +0x1163dbd)
>  #4 in os_fork_impl at workdir/UnpackedTarball/python3/./Modules/posixmodule.c:6253:9 (workdir/UnpackedTarball/python3/libpython3.8d.so.1.0 +0x11adcad)
>  #5 in os_fork at workdir/UnpackedTarball/python3/./Modules/clinic/posixmodule.c.h:2750:12 (workdir/UnpackedTarball/python3/libpython3.8d.so.1.0 +0x117b481)
>  #6 in cfunction_vectorcall_NOARGS at workdir/UnpackedTarball/python3/Objects/methodobject.c:463:24 (workdir/UnpackedTarball/python3/libpython3.8d.so.1.0 +0x9db7e1)
>  #7 in _PyObject_Vectorcall at workdir/UnpackedTarball/python3/./Include/cpython/abstract.h:127:11 (workdir/UnpackedTarball/python3/libpython3.8d.so.1.0 +0xf0225e)
>  #8 in call_function at workdir/UnpackedTarball/python3/Python/ceval.c:4963:13 (workdir/UnpackedTarball/python3/libpython3.8d.so.1.0 +0xef3f4a)
>  #9 in _PyEval_EvalFrameDefault at workdir/UnpackedTarball/python3/Python/ceval.c:3469:23 (workdir/UnpackedTarball/python3/libpython3.8d.so.1.0 +0xedc5d8)
[...]
>  #143 in pymain_run_python at workdir/UnpackedTarball/python3/Modules/main.c:610:21 (workdir/UnpackedTarball/python3/libpython3.8d.so.1.0 +0x1149f6c)
>  #144 in Py_RunMain at workdir/UnpackedTarball/python3/Modules/main.c:689:5 (workdir/UnpackedTarball/python3/libpython3.8d.so.1.0 +0x11495e9)
>  #145 in pymain_main at workdir/UnpackedTarball/python3/Modules/main.c:719:12 (workdir/UnpackedTarball/python3/libpython3.8d.so.1.0 +0x114a299)
>  #146 in Py_BytesMain at workdir/UnpackedTarball/python3/Modules/main.c:743:12 (workdir/UnpackedTarball/python3/libpython3.8d.so.1.0 +0x114a30d)
>  #147 in main at workdir/UnpackedTarball/python3/./Programs/python.c:16:12 (workdir/UnpackedTarball/python3/python +0x4d00f8)

Assuming that the GIL is always locked before the fork, better tell TSan about a
fake RELEASED before telling it about a fake DESTROY, to keep TSan's model
consistent.

Change-Id: I6c68d7e415aa0ffc3047e5a5c4c4aca6b0cce8cb
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/130985
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
2022-03-04 17:15:54 +01:00

10 lines
311 B
Text

--- Python/ceval_gil.h
+++ Python/ceval_gil.h
@@ -135,6 +135,7 @@
static void recreate_gil(struct _gil_runtime_state *gil)
{
+ _Py_ANNOTATE_RWLOCK_RELEASED(&gil->locked, 1);
_Py_ANNOTATE_RWLOCK_DESTROY(&gil->locked);
/* XXX should we destroy the old OS resources here? */
create_gil(gil);