Re: [PATCH 1/7] tests/docker: add gdb-multiarch to all-test-cross

["Alex Bennée" <alex.bennee@linaro.org>]

Ilya Leoshkevich <iii@linux.ibm.com> writes:

> On 3/12/26 14:56, Alex Bennée wrote:
>> Philippe Mathieu-Daudé <philmd@linaro.org> writes:
>>
>>> On Thu, 12 Mar 2026 at 11:39, Philippe Mathieu-Daudé <philmd@linaro.org> wrote:
>>>> On 12/3/26 11:11, Ilya Leoshkevich wrote:
>>>>> On 3/11/26 00:28, Philippe Mathieu-Daudé wrote:
>>>>>> On 11/3/26 00:22, Pierrick Bouvier wrote:
>>>>>>> On 3/10/26 4:07 PM, Philippe Mathieu-Daudé wrote:
>>>>>>>> On 10/3/26 23:33, Philippe Mathieu-Daudé wrote:
>>>>>>>>> On 6/3/26 17:17, Alex Bennée wrote:
>>>>>>>>>> This allows the gdb integration tests to be run as well.
>>>>>>>>>>
>>>>>>>>>> Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
>>>>>>>>>> ---
>>>>>>>>>>     tests/docker/dockerfiles/debian-all-test-cross.docker | 1 +
>>>>>>>>>>     tests/lcitool/refresh                                 | 1 +
>>>>>>>>>>     2 files changed, 2 insertions(+)
>>>>>>>>> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
>>>>>>>> I'm getting:
>>>>>>>>
>>>>>>>> Python Exception <class 'gdb.MemoryError'>: Cannot access memory at
>>>>>>>> address 0x45001fff
>>>>>>>> Error occurred in Python: Cannot access memory at address 0x45001fff
>>>>>>>> qemu-hppa: QEMU: Terminated via GDBstub
>>>>>>>> make[1]: ***
>>>>>>>> [/builds/philmd/qemu/tests/tcg/multiarch/Makefile.target:121:
>>>>>>>> run-gdbstub-prot-none] Error 1
>>>>>>>> make[1]: *** Waiting for unfinished jobs....
>>>>>>>> qemu-hppa: QEMU: Terminated via GDBstub
>>>>>>>> warning: Could not load shared library symbols for linux-vdso32.so.1.
>>>>>>>> Do you need "set solib-search-path" or "set sysroot"?
>>>>>>>> make: *** [/builds/philmd/qemu/tests/Makefile.include:66:
>>>>>>>> run-tcg-tests-hppa-linux-user] Error 2
>>>>>>>>
>>>>>>> To add more information, we tested it on GitHub, our machines and
>>>>>>> GitLab, and for an unknown reason, it only fails on GitLab runners.
>>>>>> Also we tested on our local Linux machines, where this test pass
>>>>>> for all our targets ;)
>>>>>>
>>>>>>> Which is unfortunate since it is where QEMU run its CI.
>>>>>>>
>>>>>>> Regards,
>>>>>>> Pierrick
>>>>> FWIW it passes on my laptop as well.
>>>>>
>>>>> Does this test fail only for hppa, or for other architectures as well?
>>>> Have a look at these job failures:
>>>> https://gitlab.com/philmd/qemu/-/pipelines/2376819801
>>>>
>>>> Python Exception <class 'gdb.MemoryError'>: Cannot access memory at
>>>> address 0x7e085b4b1fff
>>>> Error occurred in Python: Cannot access memory at address 0x7e085b4b1fff
>>>> qemu-aarch64: QEMU: Terminated via GDBstub
>>>> make[1]: ***
>>>> [/builds/philmd/qemu/tests/tcg/multiarch/Makefile.target:121:
>>>> run-gdbstub-prot-none] Error 1
>>>> make: *** [/builds/philmd/qemu/tests/Makefile.include:66:
>>>> run-tcg-tests-aarch64-linux-user] Error 2
>>>>
>>>> Python Exception <class 'gdb.MemoryError'>: Cannot access memory at
>>>> address 0x40802fff
>>>> Error occurred in Python: Cannot access memory at address 0x40802fff
>>>> qemu-arm: QEMU: Terminated via GDBstub
>>>> make[1]: ***
>>>> [/builds/philmd/qemu/tests/tcg/multiarch/Makefile.target:121:
>>>> run-gdbstub-prot-none] Error 1
>>>> make[1]: *** Waiting for unfinished jobs....
>>>> make: *** [/builds/philmd/qemu/tests/Makefile.include:66:
>>>> run-tcg-tests-arm-linux-user] Error 2
>>>>
>>>> Python Exception <class 'gdb.MemoryError'>: Cannot access memory at
>>>> address 0x45001fff
>>>> Error occurred in Python: Cannot access memory at address 0x45001fff
>>>> qemu-hppa: QEMU: Terminated via GDBstub
>>>> make[1]: ***
>>>> [/builds/philmd/qemu/tests/tcg/multiarch/Makefile.target:121:
>>>> run-gdbstub-prot-none] Error 1
>>>> make[1]: *** Waiting for unfinished jobs....
>>>> qemu-hppa: QEMU: Terminated via GDBstub
>>>> make: *** [/builds/philmd/qemu/tests/Makefile.include:66:
>>>> run-tcg-tests-hppa-linux-user] Error 2
>>>>
>>>> Python Exception <class 'gdb.MemoryError'>: Cannot access memory at
>>>> address 0x7d742025efff
>>>> Error occurred in Python: Cannot access memory at address 0x7d742025efff
>>>> qemu-alpha: QEMU: Terminated via GDBstub
>>>> make[1]: ***
>>>> [/builds/philmd/qemu/tests/tcg/multiarch/Makefile.target:121:
>>>> run-gdbstub-prot-none] Error 1
>>>> make: *** [/builds/philmd/qemu/tests/Makefile.include:66:
>>>> run-tcg-tests-alpha-linux-user] Error 2
>>>>
>>>> So at least Alpha / ARM / Aarch64 / HPPA.
>>>>
>>>> (I haven't checked whether the test is run for other targets).
>>>>
>>>> We also run these on GitHub runners where the tests pass, and apparently
>>>> when running on k8s private runner on GitLab they also pass. The problem
>>>> is with public runner on GitLab.
>>> Also another test we ran was to pull GitLab docker image and run the same
>>> tests locally within that GitLab-generated image and they passed. So we
>>> believe the issue is related to some container restriction with
>>> GitLab.
>> Most likely because the ptrace settings are tighter - ptrace is the
>> mechanism we use to access the memory we can't normally.
>
>
> I thought about this, but then probe_proc_self_mem() that guards the
> test would fail.
>
> I don't think cpu_memory_rw_debug() does anything in addition to what
> we probe for.

The probe only checks for read access. Could there be a difference
between read and write?
>
>
> [...]

-- 
Alex Bennée
Virtualisation Tech Lead @ Linaro