python -m pdb -p fails when CONFIG_CROSS_MEMORY_ATTACH not set in kernel config

[cakemanny]

Bug report

Bug description:

It's possible that a Linux kernel is built with the kernel config variable CONFIG_CROSS_MEMORY_ATTACH not set. (This is the case for the kernel that comes with Flatcar Container Linux and for CoreOS, the former of which is used for kubernetes nodes by the cloud provider we use).

In this case, using python -m pdb <some python PID> fails and it looks like this:

root@bc6b29d0894d:/src/cpython# ./python -c $'import time;\nwhile True: time.sleep(1)' &
[1] 17106
root@bc6b29d0894d:/src/cpython# ./python -m pdb -p $!
OSError: [Errno 38] Function not implemented

The above exception was the direct cause of the following exception:

OSError: process_vm_readv failed for PID 17106 at address 0xaaaab1f8d038 (size 760, partial read 0 bytes): Function not implemented

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/src/cpython/Lib/runpy.py", line 198, in _run_module_as_main
    return _run_code(code, main_globals, None,
                     "__main__", mod_spec)
  File "/src/cpython/Lib/runpy.py", line 88, in _run_code
    exec(code, run_globals)
    ~~~~^^^^^^^^^^^^^^^^^^^
  File "/src/cpython/Lib/pdb.py", line 3609, in <module>
    pdb.main()
    ~~~~~~~~^^
  File "/src/cpython/Lib/pdb.py", line 3540, in main
    attach(opts.pid, opts.commands)
    ~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/src/cpython/Lib/pdb.py", line 3424, in attach
    sys.remote_exec(pid, connect_script.name)
    ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Failed to read debug offsets structure from remote process

This is a single command to repro on such a system

docker run -i --rm python:3.13 bash << 'EOF'  # the python version here is irrelevant, the image just has the right deps
newresolv=$(sed $'/nameserver/i\\\nnameserver 1.1.1.1' /etc/resolv.conf); echo "$newresolv" > /etc/resolv.conf  # resolv is wrong in my repro VM
mkdir src && cd src
git clone https://github.com/python/cpython.git --depth=10  # add --branch=3.14 to test 3.14
cd cpython
./configure --with-pydebug && make -j $(nproc)
./python -c $'import time;\nwhile True: time.sleep(1)' &
./python -m pdb -p $!
EOF

Further relevant version info for the above repro output:

root@bc6b29d0894d:/src/cpython# uname -a
Linux bc6b29d0894d 6.6.88-flatcar #1 SMP PREEMPT Tue Apr 29 23:08:45 -00 2025 aarch64 GNU/Linux

root@bc6b29d0894d:/src/cpython# ./python -VV
Python 3.15.0a0 (heads/main:e64395e, May 29 2025, 09:16:22) [GCC 12.2.0]

root@bc6b29d0894d:/src/cpython# git show | head -1
commit e64395e8eb8d3a9e35e3e534e87d427ff27ab0a5

root@bc6b29d0894d:/src/cpython# zcat /proc/config.gz  | grep CONFIG_CROSS_MEMORY_ATTACH
# CONFIG_CROSS_MEMORY_ATTACH is not set

Diagnosis:
When CONFIG_CROSS_MEMORY_ATTACH is not set, the syscall handler for process_vm_readv (and also process_vm_writev) is not compiled into the kernel and so it returns ENOSYS as errno.

If it's ok, I'd also like to / be happy to offer a PR to fix.
Specifically I'd go with the approach that I discovered someone contributed to py-spy, of falling back to the /proc/[pid]/mem file in the proc filesystem. I could even open such a PR pretty soon.

CPython versions tested on:

3.14, 3.15, CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-134876: Add fallback for when process_vm_readv fails with ENOSYS #134878

[ZeroIntensity]

(Hit the assign button on accident)