Repository navigation
test_manual_unwind_respects_frame_pointers fails on x86_64 Fedora Linux with enabled JIT (yes-off) #144741
Description
Bug report
Bug description:
We build Python on x86_64 and aarch64 with --enable-experimental-jit=yes-off . The newly added test passes for aarch64 but doesn't for x86_64, with the following traceback.
For completeness, the CFLAGS we build with are:
CFLAGS=' -flto=auto -ffat-lto-objects -fexceptions -g -grecord-gcc-switches -pipe -Wall -Wno-complain-wrong-lang -Werror=format-security -Wp,-U_FORTIFY_SOURCE,-D_FORTIFY_SOURCE=3 -Wp,-D_GLIBCXX_ASSERTIONS -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -fstack-protector-strong -specs=/usr/lib/rpm/redhat/redhat-annobin-cc1 -m64 -march=x86-64 -mtune=generic -fasynchronous-unwind-tables -fstack-clash-protection -fcf-protection -mtls-dialect=gnu2 -fno-omit-frame-pointer -mno-omit-leaf-frame-pointer 'I am unable to reproduce this outside of the rpm build context yet, but I'm happy to test any ideas.
Full traceback:
Re-running test_frame_pointer_unwind in verbose mode (matching: test_manual_unwind_respects_frame_pointers)
test_manual_unwind_respects_frame_pointers (test.test_frame_pointer_unwind.FramePointerUnwindTests.test_manual_unwind_respects_frame_pointers) ... #00 0x7f8763378fda -> other
#01 0x7f87632eeac7 -> other
#02 0x7f87632efb77 -> other
#03 0x7f87634ddd3c -> other
#04 0x7f87634e6604 -> other
#05 0x7f87634da121 -> other
#06 0x7f876351552d -> other
#07 0x7f87632eff85 -> other
#08 0x7f87632eeac7 -> other
#09 0x7f87632efb77 -> other
#10 0x7f8763852186 -> other
#11 0x7f87634de1f1 -> other
#12 0x7f87634e8cdc -> other
#13 0x7f87634da121 -> other
#14 0x7f876351552d -> other
#15 0x7f87632eff85 -> other
#16 0x7f87632eeac7 -> other
#17 0x7f87632efb77 -> other
#18 0x7f8763852186 -> other
#19 0x7f87634de1f1 -> other
#20 0x7f87634e8cdc -> other
#21 0x7f87634da121 -> other
#22 0x7f876351552d -> other
#23 0x7f87632eff85 -> other
#24 0x7f87632eeac7 -> other
#25 0x7f87632efb77 -> other
#26 0x7f8763852186 -> other
#27 0x7f87634de1f1 -> other
#28 0x7f87634e8cdc -> other
#29 0x7f87634da121 -> other
#30 0x7f876351552d -> other
#31 0x7f87632eff85 -> other
#32 0x7f87632eeac7 -> other
#33 0x7f87632efb77 -> other
#34 0x7f8763852186 -> other
#35 0x7f87634de1f1 -> other
#36 0x7f87634e8cdc -> other
#37 0x7f87634da121 -> other
#38 0x7f876351552d -> other
#39 0x7f87632eff85 -> other
#40 0x7f87632eeac7 -> other
#41 0x7f87632efb77 -> other
#42 0x7f8763852186 -> other
#43 0x7f87634de1f1 -> other
#44 0x7f87634e8cdc -> other
#45 0x7f87634da121 -> other
#46 0x7f876351552d -> other
#47 0x7f87632eff85 -> other
#48 0x7f87632eeac7 -> other
#49 0x7f87632efb77 -> other
#50 0x7f8763852186 -> other
#51 0x7f87634de1f1 -> other
#52 0x7f87634e8cdc -> other
#53 0x7f87634da121 -> other
#54 0x7f876351552d -> other
#55 0x7f87632eff85 -> other
#56 0x7f87632eeac7 -> other
#57 0x7f87632efb77 -> other
#58 0x7f8763852186 -> other
#59 0x7f87634de1f1 -> other
#60 0x7f87634e8cdc -> other
#61 0x7f87634da121 -> other
#62 0x7f876351552d -> other
#63 0x7f87632eff85 -> other
#64 0x7f87632eeac7 -> other
#65 0x7f87632efb77 -> other
#66 0x7f8763852186 -> other
#67 0x7f87634de1f1 -> other
#68 0x7f87634e8cdc -> other
#69 0x7f87634da121 -> other
#70 0x7f876351552d -> other
#71 0x7f87632eff85 -> other
#72 0x7f87632eeac7 -> other
#73 0x7f87632efb77 -> other
#74 0x7f8763852186 -> other
#75 0x7f87634de1f1 -> other
#76 0x7f87634e8cdc -> other
#77 0x7f87634da121 -> other
#78 0x7f876351552d -> other
#79 0x7f87632eff85 -> other
#80 0x7f87632eeac7 -> other
#81 0x7f87632efb77 -> other
#82 0x7f8763852186 -> other
#83 0x7f8763378e09 -> other
#84 0x7f87632eeac7 -> other
#85 0x7f87632efb77 -> other
#86 0x7f87634ddd3c -> other
#87 0x7f87634e6604 -> other
#88 0x7f87634da121 -> other
#89 0x7f876351552d -> other
#90 0x7f87634dd758 -> other
#91 0x7f876374d7ed -> other
#92 0x7f876374dc58 -> other
#93 0x7f876374d40e -> other
#94 0x7f876374b98d -> other
#95 0x7f876378b0fd -> other
#96 0x7f876378c121 -> other
#97 0x7f876378c2f3 -> other
#98 0x7f876378c3c9 -> other
#99 0x7f876378c491 -> other
#100 0x55f0d60984bd -> python
#101 0x7f8763008681 -> other
#102 0x7f8763008798 -> other
#103 0x55f0d60983d5 -> python
{"length": 104, "python_frames": 2, "jit_frames": 0, "other_frames": 102, "jit_backend": "jit"}
JIT ranges:
0x7fb224624000-0x7fb224647000
#00 0x7fb233378fda -> other
#01 0x7fb2332eeac7 -> other
#02 0x7fb2332efb77 -> other
#03 0x7fb2334ddd3c -> other
#04 0x7fb2334e6604 -> other
#05 0x7fb2334da121 -> other
#06 0x7fb23351552d -> other
#07 0x7fb2332eff85 -> other
#08 0x7fb2332eeac7 -> other
#09 0x7fb2332efb77 -> other
#10 0x7fb233852186 -> other
#11 0x7fb2334de1f1 -> other
#12 0x7fb22462f00f -> jit
{"length": 13, "python_frames": 0, "jit_frames": 1, "other_frames": 12, "jit_backend": "jit"}
test_manual_unwind_respects_frame_pointers (test.test_frame_pointer_unwind.FramePointerUnwindTests.test_manual_unwind_respects_frame_pointers) (env={'PYTHON_JIT': '1'}) ... FAIL
======================================================================
FAIL: test_manual_unwind_respects_frame_pointers (test.test_frame_pointer_unwind.FramePointerUnwindTests.test_manual_unwind_respects_frame_pointers) (env={'PYTHON_JIT': '1'})
----------------------------------------------------------------------
Traceback (most recent call last):
File "/builddir/build/BUILD/python3.15-3.15.0_a6-build/Python-3.15.0a6/Lib/test/test_frame_pointer_unwind.py", line 182, in test_manual_unwind_respects_frame_pointers
self.assertGreater(
~~~~~~~~~~~~~~~~~~^
python_frames,
^^^^^^^^^^^^^^
0,
^^
f"expected to find Python frames on {self.machine} with env {env}",
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
AssertionError: 0 not greater than 0 : expected to find Python frames on x86_64 with env {'PYTHON_JIT': '1'}
CPython versions tested on:
3.15
Operating systems tested on:
Linux
Activity
vstinner commentedon Mar 4, 2026
I reproduced the failure on Fedora Rawhide with the following commands:
Fedora Rawhide: gcc (GCC) 16.0.1 20260209 (Red Hat 16.0.1-0).
vstinner commentedon Mar 4, 2026
Aha, I can also reproduce the failure on Fedora 43: gcc (GCC) 15.2.1 20260123 (Red Hat 15.2.1-7).
vstinner commentedon Mar 4, 2026
I reduced the reproducer to:
vstinner commentedon Mar 4, 2026
script.py:Python built with
--enable-sharedOutput with JIT disabled (truncated):
python_frames=2Output with JIT enabled:
python_frames=0Python built without
--enable-sharedOutput with JIT disabled (truncated):
python_frames=104Output with JIT enabled:
python_frames=12pythongh-144741: Fix test_frame_pointer_unwind for libpython
pythongh-144741: Fix test_frame_pointer_unwind for libpython
vstinner commentedon Mar 4, 2026
I wrote #145499 to fix this issue.
gh-144741: Fix test_frame_pointer_unwind for libpython (#145499)
vstinner commentedon Mar 4, 2026
Fixed by the change a003923. Thanks for your bug report @befeleme.
ferdnyc commentedon Mar 10, 2026
Unfortunately, even after #145499 I'm seeing
test_frame_pointer_unwindsegfaults on Fedora 43 x86_64 after building Python using system GCC 15.2.1 (wrapped with automatic ccache), in an out-of-tree builddir (in case that's significant) configured with--enable-shared. Exact commands used:Output for failing test:
hroncok commentedon Mar 10, 2026
I was about to say that 3.15.0a7 indeed fixed this for us. Will try on Fedora 43 as well.
ferdnyc commentedon Mar 10, 2026
Running the individual test through
gdb, the segfault seems to happen in the new_testinternalcapi.cfunctionmanual_unwind_from_fp, when it gets passed a very suspiciousframe_pointervalue:(
return_addris set toframe_pointer[1]a few lines earlier, but given that it appears to be no sort of pointer..)Here's the start of the backtrace, up to the point it gets ugly and redundant. Not that it's very useful with the arguments to
manual_frame_pointer_unwindoptimized out:(Is
nargsfsupposed to be an argument count? Because if so, seems like something is already way wrong.)ferdnyc commentedon Mar 10, 2026
I'm running on
x86_64, as I said, building without anyomit-frame-pointerflags in theCFLAGS. As a result, when the test'ssetUpmethod runs_frame_pointers_expected("x86_64"), it gets backFalse, as expected. (Confirmed by manually defining and executing the function in the REPL.)It then executes this:
...Which is where it segfaults.
I assume if I was building with
-fno-omit-frame-pointer(like the Fedora official builds are, according to @befeleme's OP) then it wouldn't be an issue, but my build doesn't have that compile flag set.ferdnyc commentedon Mar 11, 2026
Huh. Stepping through with
gdb, I'm inside amanual_unwind_from_fp()call that was passed a valid ("valid-looking")frame_pointer=0x7fffffffb6f0. But here's what I get if I pick that value apart roughly the way the code does:So, that's where the crazy-looking
frame_pointer=0x8000000000000000is coming from; it's when the code gets to the second iteration of this loop, becausenext_fpis apparently neitherNULL, nor less thanmin_frame_pointer_addr, nor a valid frame pointer...Methinks something is not the size/shape we think it is. (The return type from
__builtin_frame_address()isvoid*, after all.)cpython/Modules/_testinternalcapi.c
Lines 345 to 384 in 5197ecb
vstinner commentedon Mar 11, 2026
Please open a new issue for your crash.
vstinner commentedon Mar 11, 2026
Thanks for the confirmation, I close the issue.
Update python315 to version 3.15.0~a7 / rev 8 via SR 1341615
mcepl commentedon Apr 8, 2026
What’s the relationship between this and #126910 ?
vstinner commentedon Apr 8, 2026
This issue is about a failure of
test_frame_pointer_unwind.test_frame_pointer_unwind,_testinernalcapi.manual_frame_pointer_unwind()and_testinternalcapi.classify_stack_addresses()were added by the issue GH-126910 (commit 66055d0).pythongh-144741: Fix test_frame_pointer_unwind for libpython (python#…