Re: [Qemu-devel] [PATCH] iothread: fix iothread hang when stop too soon

[Thomas Huth]

On 2019-01-29 06:14, Peter Xu wrote:
> Lukas reported an hard to reproduce QMP iothread hang on s390 that
> QEMU might hang at pthread_join() of the QMP monitor iothread before
> quitting:
> 
>   Thread 1
>   #0  0x000003ffad10932c in pthread_join
>   #1  0x0000000109e95750 in qemu_thread_join
>       at /home/thuth/devel/qemu/util/qemu-thread-posix.c:570
>   #2  0x0000000109c95a1c in iothread_stop
>   #3  0x0000000109bb0874 in monitor_cleanup
>   #4  0x0000000109b55042 in main
> 
> While the iothread is still in the main loop:
> 
>   Thread 4
>   #0  0x000003ffad0010e4 in ??
>   #1  0x000003ffad553958 in g_main_context_iterate.isra.19
>   #2  0x000003ffad553d90 in g_main_loop_run
>   #3  0x0000000109c9585a in iothread_run
>       at /home/thuth/devel/qemu/iothread.c:74
>   #4  0x0000000109e94752 in qemu_thread_start
>       at /home/thuth/devel/qemu/util/qemu-thread-posix.c:502
>   #5  0x000003ffad10825a in start_thread
>   #6  0x000003ffad00dcf2 in thread_start
> 
> IMHO it's because there's a race between the main thread and iothread
> when stopping the thread in following sequence:
> 
>     main thread                       iothread
>     ===========                       ==============
>                                       aio_poll()
>     iothread_get_g_main_context
>       set iothread->worker_context
>     iothread_stop
>       schedule iothread_stop_bh
>                                         execute iothread_stop_bh [1]
>                                           set iothread->running=false
>                                           (since main_loop==NULL so
>                                            skip to quit main loop.
>                                            Note: although main_loop is
>                                            NULL but worker_context is
>                                            not!)
>                                       atomic_read(&iothread->worker_context) 
> [2]
>                                         create main_loop object
>                                         g_main_loop_run() [3]
>     pthread_join() [4]
> 
> We can see that when execute iothread_stop_bh() at [1] it's possible
> that main_loop is still NULL because it's only created until the first
> check of the worker_context later at [2].  Then the iothread will hang
> in the main loop [3] and it'll starve the main thread too [4].
> 
> Here the simple solution should be that we check again the "running"
> variable before check against worker_context.
> 
> CC: Thomas Huth <address@hidden>
> CC: Dr. David Alan Gilbert <address@hidden>
> CC: Stefan Hajnoczi <address@hidden>
> CC: Lukáš Doktor <address@hidden>
> CC: Markus Armbruster <address@hidden>
> CC: Eric Blake <address@hidden>
> CC: Paolo Bonzini <address@hidden>
> Reported-by: Lukáš Doktor <address@hidden>
> Signed-off-by: Peter Xu <address@hidden>
> ---
> 
> This hasn't yet been verified on the initial s390 systems, but since I
> can reproduce it locally with this code clip:
> 
>         IOThread *iothread = iothread_create("test", NULL);
>         iothread_get_g_main_context(iothread);
>         iothread_stop(iothread);
> 
> so I'm still posting this out for review first in case it was hit by
> other users.
> ---
>  iothread.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/iothread.c b/iothread.c
> index 2fb1cdf55d..e615b7ae52 100644
> --- a/iothread.c
> +++ b/iothread.c
> @@ -63,7 +63,11 @@ static void *iothread_run(void *opaque)
>      while (iothread->running) {
>          aio_poll(iothread->ctx, true);
>  
> -        if (atomic_read(&iothread->worker_context)) {
> +        /*
> +         * We must check the running state again in case it was
> +         * changed in previous aio_poll()
> +         */
> +        if (iothread->running && atomic_read(&iothread->worker_context)) {
>              GMainLoop *loop;
>  
>              g_main_context_push_thread_default(iothread->worker_context);
> 

I ran this on s390x with Lukáš' reproducer for a while now, and so far I
haven't seen any hangs anymore. Thus this seems to fix the issue as far
as I can see, thanks!

Tested-by: Thomas Huth <address@hidden>