On Sun, 15 Jul 2018 14:42:23 +0200
Paolo Bonzini <pbonz...@redhat.com> wrote:
> On 13/07/2018 20:42, Stefan Hajnoczi wrote:
> > +#ifndef _WIN32
> > +static void stop_writeout_thread(void)
> > +{
> > + g_mutex_lock(&trace_lock);
> > + trace_writeout_running = false;
> > + g_cond_signal(&trace_available_cond);
> > + g_mutex_unlock(&trace_lock);
> > +
> > + g_thread_join(trace_writeout_thread);
> > + trace_writeout_thread = NULL;
> > +}
>
> After stop_writeout_thread returns, another could start a write to the
> shared data structure---and the write would never finish, because the
> thread disappears after fork(2) returns. This would leave the mutex
> locked, causing a deadlock soon after the fork. So you need to lock
> trace_lock again here, and unlock it in restart_writeout_thread.
So, I suppose there will be a v3, right?
>
> Apart from this, it looks good!
Did a quick run and it fixed the problems for me.
