Re: [PATCH V12 06/15] monitor: track IOThread users with QOM paths

On Thu, Aug 20, 2026 at 3:12 AM Stefan Hajnoczi <[email protected]> wrote:
>
> On Sat, Aug 15, 2026 at 01:48:20AM +0800, Zhang Chen wrote:
> > @@ -668,6 +668,17 @@ void monitor_cleanup(void)
> >          qemu_mutex_unlock(&monitor_lock);
> >          monitor_flush(mon);
> >          qemu_mutex_lock(&monitor_lock);
> > +
> > +        if (monitor_requires_iothread(mon)) {
> > +            g_autofree char *path = object_get_canonical_path(OBJECT(mon));
> > +            const IOThreadHolder io_holder = {
> > +                .type = IO_THREAD_HOLDER_KIND_QOM_OBJECT,
> > +                .u.qom_object.qom_path = path,
> > +            };
> > +
> > +            iothread_unref_and_put_aio_context(mon_iothread, &io_holder);
> > +            mon->ctx = NULL;
> > +        }
>
> Is it possible to move this to monitor_finalize() so it's guaranteed to
> run when a Monitor object is deleted with object_unparent()? That would
> avoid the code duplication in monitor_qmp_prepare_delete().
>
> One complication is that object_get_canonical_path() may not work inside
> monitor_finalize(), so it may be necessary to store the IOThreadHolder
> or at least the qom_path in Monitor.

Yes. Since the canonical path is no longer available in monitor_finalize(),
I’ll move the cleanup to the Monitor’s ObjectClass::unparent callback instead.
It runs before the object is detached, avoids storing the path, and removes the
duplication in monitor_cleanup() and monitor_qmp_prepare_delete().

Thanks
Chen