Re: [PATCH 14/19] util/async: Fixed tsan warnings

[Robert Foley]

Hi Stefan,

On Tue, 26 May 2020 at 06:32, Stefan Hajnoczi <address@hidden> wrote:
>
> On Fri, May 22, 2020 at 12:07:50PM -0400, Robert Foley wrote:
> > For example:
> >   Atomic write of size 8 at 0x7b4800113c28 by main thread (mutexes: write 
> > M30):
> >     #0 __tsan_atomic64_exchange <null> (qemu-system-aarch64+0x386f85)
> >     #1 aio_bh_poll util/async.c:146:5 (qemu-system-aarch64+0xcd1f61)
> >     #2 aio_dispatch util/aio-posix.c:380:5 (qemu-system-aarch64+0xcd8abb)
> >     #3 aio_ctx_dispatch util/async.c:298:5 (qemu-system-aarch64+0xcd31b0)
> >     #4 g_main_context_dispatch <null> (libglib-2.0.so.0+0x4c416)
> >     #5 qemu_main_loop softmmu/vl.c:1664:9 (qemu-system-aarch64+0x5cc6d6)
> >     #6 main softmmu/main.c:49:5 (qemu-system-aarch64+0xc62857)
> >
> >   Previous read of size 8 at 0x7b4800113c28 by thread T3 (mutexes: write 
> > M81):
> >     #0 aio_bh_enqueue util/async.c:81:9 (qemu-system-aarch64+0xcd2267)
> >     #1 qemu_bh_schedule util/async.c:181:5 (qemu-system-aarch64+0xcd2267)
> >     #2 worker_thread util/thread-pool.c:113:9 (qemu-system-aarch64+0xcd473c)
> >     #3 qemu_thread_start util/qemu-thread-posix.c:519:9 
> > (qemu-system-aarch64+0xcde280)
> >
> > Cc: Stefan Hajnoczi <address@hidden>
> > Cc: Fam Zheng <address@hidden>
> > Signed-off-by: Robert Foley <address@hidden>
> > ---
> >  util/async.c | 11 +++++++++--
> >  1 file changed, 9 insertions(+), 2 deletions(-)
>
> Please explain why these warnings should be ignored in the commit
> description. I have been CCed on this patch, am unfamiliar with TSAN
> annotations, and now need to figure out:
> 1. Is ignoring the warning safe or should async.c be fixed somehow?
> 2. How do the annotation macros work and are they being used correectly?
>
> It's likely that people coming across this commit in the future would
> also benefit from information in the commit description that makes these
> things clear.
>
All good points.

I think we need to add a whole lot more details on TSan and its use
along with the use of the annotation macros among other things.
Will CC you on the updates to testing.rst, which will contain this
information.
This will hopefully provide the context, information and resources needed to
make the kinds of determinations we're looking for in 1. and 2.

As for the annotations, I think we are going to re-focus the patch and
remove the annotations for now.  However, we will plan to update
this commit description with more details on the changes to use atomic reads.

Thanks & Regards,
-Rob
> Thanks,
> Stefan