Re: Thread safety of coroutine-sigaltstack

[Laszlo Ersek <lersek@redhat.com>]

On 01/22/21 11:14, Peter Maydell wrote:
> On Fri, 22 Jan 2021 at 08:50, Max Reitz <mreitz@redhat.com> wrote:
>>
>> On 20.01.21 18:25, Laszlo Ersek wrote:
>>
>> [...]
>>
>>> A simple grep for SIGUSR2 seems to indicate that SIGUSR2 is not used by
>>> system emulation for anything else, in practice. Is it possible to
>>> dedicate SIGUSR2 explicitly to coroutine-sigaltstack, and set up the
>>> action beforehand, from some init function that executes on a "central"
>>> thread, before qemu_coroutine_new() is ever called?
>>
>> I wrote a patch to that effect, but just before sending I wondered
>> whether SIGUSR2 cannot be registered by the “guest” in user-mode
>> emulation, and whether that would then break coroutines from there on.
>>
>> (I have no experience dealing with user-mode emulation, but it does look
>> like the guest can just register handlers for any signal but SIGSEGV and
>> SIGBUS.)
> 
> Yes, SIGUSR2 is for the guest in user-emulation mode. OTOH do we
> even use the coroutine code in user-emulation mode? Looking at
> the meson.build files, we only add the coroutine_*.c to util_ss
> if 'have_block', and we set have_block = have_system or have_tools.
> I think (but have not checked) that that means we will build and
> link the object file into the user-mode binaries if you happen
> to build them in the same run as system-mode binaries,

I did that, first running

 ./configure \
    --enable-debug \
    --target-list==x86_64-softmmu,x86_64-linux-user \
    --with-coroutine=sigaltstack

Then I checked the "qemu-system-x86_64" and "qemu-x86_64" binaries with
"nm". Only the former contains "coroutine_init":

00000000009725e4 t coroutine_init

So I believe the coroutine object file(s) are not even linked into the
user-mode emulators. (coroutine_init() is a constructor function, so I
think it would be preserved otherwise, even if it had no explicit caller.)

I tried a different approach too: an #error in
"coroutine-sigaltstack.c", if CONFIG_LINUX_USER were #defined. But that
aborted the build, due to CONFIG_LINUX_USER being poisoned in the first
place. Maybe that result was already enough to answer the question, but
I wasn't sure, hence the check with "nm".

Thanks,
Laszlo

> but won't
> build them in if you built the user-mode binaries as a separate
> build. Which is odd and probably worth fixing, but does mean we
> know that we aren't actually using coroutines in user-mode.
> (Also user-mode really means Linux or BSD and I think both of
> those have working ucontext.)
> 
> thanks
> -- PMM
>