Native unwinding fails across signal frames on AArch64

[K0u1e]

Is there an existing issue for this?

• I have searched the existing issues

Current Behavior

It looks like elfutils fails to unwind past the signal frame on AArch64, so pystack cannot recover the native frames below it. Would you be interested in a patch for this?

Expected Behavior

No response

Steps To Reproduce

run this script

import ctypes
import os
import signal
import subprocess


@ctypes.CFUNCTYPE(None, ctypes.c_int)
def dump_stack(*_):
    subprocess.run(
        ["pystack", "remote", "--native", "--no-color", str(os.getpid())]
    )


libc = ctypes.CDLL(None)
libc.signal.argtypes = (ctypes.c_int, ctypes.c_void_p)
libc.signal(signal.SIGUSR1, dump_stack)
os.kill(os.getpid(), signal.SIGUSR1)

and get

* - Unable to merge native stack due to insufficient native information - *

Pystack Version

1.7.0

Python Version

3.13

Linux distribution

Debian

Anything else?

No response

[godlygeek]

What patch do you propose? This seems more like something that should be patched in elfutils than in pystack, at first glance...

[K0u1e]

What patch do you propose? This seems more like something that should be patched in elfutils than in pystack, at first glance...

Yes, a patch in elfutils to recognize rt_sigreturn signal frames on AArch64, similar to the implementation in libunwind:
https://github.com/libunwind/libunwind/blob/master/src/aarch64/Gis_signal_frame.c

I have already reported this issue to elfutils: https://sourceware.org/bugzilla/show_bug.cgi?id=34472, but there has been no response so far.

Maybe we can apply this patch when building the pystack wheel as a workaround.

[godlygeek]

I have already reported this issue to elfutils: https://sourceware.org/bugzilla/show_bug.cgi?id=34472, but there has been no response so far.

Hm. The elfutils report doesn't have a reproducer, which makes it much harder to triage than this one is. I'd strongly recommend updating that bug report with a reproducer.

I'd expect that this would reproduce the issue:

import ctypes
import os
import signal
import subprocess


@ctypes.CFUNCTYPE(None, ctypes.c_int)
def dump_stack(*_):
    subprocess.run(
        ["eu-stack", "-p", str(os.getpid())]
    )


libc = ctypes.CDLL(None)
libc.signal.argtypes = (ctypes.c_int, ctypes.c_void_p)
libc.signal(signal.SIGUSR1, dump_stack)
os.kill(os.getpid(), signal.SIGUSR1)

(after a sudo sysctl kernel.yama.ptrace_scope=0).

---

I am willing to carry a local patch to elfutils if it's small enough, but obviously it's much better if this gets fixed in elfutils so that everyone benefits from the fix and not just pystack.