Re: [PATCH 20/20] target/loongarch: Add linux-user emulation support

[Alex Bennée]

Thomas Huth <thuth@redhat.com> writes:

> On 01/07/2021 03.07, maobibo wrote:
>> 在 2021年06月30日 17:36, Alex Bennée 写道:
>>>
>>> maobibo <maobibo@loongson.cn> writes:
>>>
>>>> 在 2021年06月29日 21:42, Peter Maydell 写道:
>>>>> On Mon, 28 Jun 2021 at 13:05, Song Gao <gaosong@loongson.cn> wrote:
>>>>>>
>>>>>> Add files to linux-user/loongarch64
>>>>>> Add file to default-configs
>>>>>> Add loongarch to target/meson.build
>>>>>>
>>>>>> Signed-off-by: Song Gao <gaosong@loongson.cn>
>>>>>> ---
>>>>>>   MAINTAINERS                                        |   1 +
>>>>>>   default-configs/targets/loongarch64-linux-user.mak |   4 +
>>>>>>   include/elf.h                                      |   2 +
>>>>>>   linux-user/elfload.c                               |  58 ++++
>>>>>>   linux-user/loongarch64/cpu_loop.c                  | 177 ++++++++++++
>>>>>>   linux-user/loongarch64/signal.c                    | 193 +++++++++++++
>>>>>>   linux-user/loongarch64/sockbits.h                  |   1 +
>>>>>>   linux-user/loongarch64/syscall_nr.h                | 307 
>>>>>> +++++++++++++++++++++
>>>>>>   linux-user/loongarch64/target_cpu.h                |  36 +++
>>>>>>   linux-user/loongarch64/target_elf.h                |  14 +
>>>>>>   linux-user/loongarch64/target_fcntl.h              |  12 +
>>>>>>   linux-user/loongarch64/target_signal.h             |  28 ++
>>>>>>   linux-user/loongarch64/target_structs.h            |  49 ++++
>>>>>>   linux-user/loongarch64/target_syscall.h            |  46 +++
>>>>>>   linux-user/loongarch64/termbits.h                  | 229 
>>>>>> +++++++++++++++
>>>>>>   linux-user/syscall_defs.h                          |   8 +-
>>>>>>   meson.build                                        |   2 +-
>>>>>>   qapi/machine-target.json                           |   4 +-
>>>>>>   target/loongarch/meson.build                       |  19 ++
>>>>>>   target/meson.build                                 |   1 +
>>>>>>   20 files changed, 1185 insertions(+), 6 deletions(-)
>>>>>
>>>>> This is a massive patch that would benefit from being split up
>>>>> into multiple smaller patches.
>>>>>
>>>>> I'm told by a kernel developer that loongarch hasn't yet been
>>>>> accepted into the Linux kernel mainline. Until it has been, the
>>>>> syscall ABI for it is not yet stable, so we won't be able to take
>>>>> the linux-user patches for it yet. (We have been burned in the
>>>>> past by taking linux-user architecture support patches without
>>>>> realizing they weren't for a stable ABI, and then being out of
>>>>> sync with the eventual upstream kernel ABI that was accepted.)
>>>>>
>>>>> We can certainly do code review in the meantime, though.
>>>> Thanks for reviewing the big series patches. It is understandable that
>>>> there should be linux kernel merged for one new architecture support
>>>> firstly, and then there will be linux-user simulator later.
>>>>
>>>> We are planning to submit patch to linux kernel for LoongArch support,
>>>> there is the link:
>>>> https://git.kernel.org/pub/scm/linux/kernel/git/chenhuacai/linux-loongson.git/log/?h=loongarch-next
>>>>
>>>> And we will continueto  submit softmmu support for LoongArch. And is there
>>>> any extra requirements for softmmu simulation for new architecture,
>>>> such as gcc/binutils/bios?
>>>
>>> Ideally if there are some pre-built toolchains either as part of a
>>> distro (we've used Debian Sid before for some) or easily to install in a
>>> docker container as binary tarballs (like we do for tricore) then we can
>>> enable basic check-tcg functionality.
>>>
>>> Going forward having stable URLs for test images of distros means we can
>>> also enable check-acceptance tests.
>> Thanks for guidance, it requires that linux kernel/gcc/glibc are submitted
>> already. My meaning is that linux-user emulation depends on kernel
>> syscall ABI, softmmu emulation does not have such dependency, on the contrast
>> system emulation can be used to verify linux kernel. Is there any requirement
>> for system emulation of new architecture?
>
> I think we'd need some way of automatic regression testing for new
> targets. If your board has a serial UART that is easy to use, then
> please add a test in tests/qtest/boot-serial-test.c.

You can also write a very minimal softmmu harness with either
semihosting or a debug port - currently we can run the tests for
x86/arm/alpha. See tests/tcg/multiarch/system and the various
boot.S/kernel.ld in tests/tcg/FOO/system and the Makefile.softmmu-target
in the relevant directories.

Basically you need enough initialisation to get to a non-stdlib main and
a __sys_outc and __exit implementation. This allows for simple smoke
tests that don't need a full Linux/kernel to run.

> And if there are
> any public Linux kernel binaries for (reliable) download anywhere,
> please add a test in tests/acceptance/, see for example the various
> machine_*.py files or boot_linux_console.py there.
>
>  HTH,
>   Thomas


-- 
Alex Bennée