Re: [PATCH V12 13/15] monitor/hmp: display IOThread holders

Zhang Chen <[email protected]> writes:

> On Thu, Aug 20, 2026 at 4:17 PM Markus Armbruster <[email protected]> wrote:
>>
>> Zhang Chen <[email protected]> writes:
>>
>> > Display holder information in info iothreads to aid manual debugging.
>> > Omit the holders line when an IOThread has no holders.
>> >
>> > Signed-off-by: Zhang Chen <[email protected]>
>> > ---
>> >  monitor/hmp-cmds.c | 21 +++++++++++++++++++++
>> >  1 file changed, 21 insertions(+)
>> >
>> > diff --git a/monitor/hmp-cmds.c b/monitor/hmp-cmds.c
>> > index e9fb8d827a..41a813e79f 100644
>> > --- a/monitor/hmp-cmds.c
>> > +++ b/monitor/hmp-cmds.c
>> > @@ -240,11 +240,32 @@ void hmp_info_iothreads(Monitor *mon, const QDict *qdict)
>> >      IOThreadInfoList *info_list = qmp_query_iothreads(NULL);
>> >      IOThreadInfoList *info;
>> >      IOThreadInfo *value;
>> > +    IOThreadHolderList *h;
>> >
>> >      for (info = info_list; info; info = info->next) {
>> >          value = info->value;
>> >          monitor_printf(mon, "%s:\n", value->id);
>> >          monitor_printf(mon, "  thread_id=%" PRId64 "\n", value->thread_id);
>> > +        monitor_printf(mon, "  holders=");
>> > +        for (h = value->holders; h; h = h->next) {
>> > +            IOThreadHolder *holder = h->value;
>> > +
>> > +            switch (holder->type) {
>> > +            case IO_THREAD_HOLDER_KIND_BLOCK_EXPORT:
>> > +                monitor_printf(mon, "[block-export: %s]",
>> > +                               holder->u.block_export.export_id);
>> > +                break;
>> > +            case IO_THREAD_HOLDER_KIND_QOM_OBJECT:
>> > +                monitor_printf(mon, "[qom-path: %s]",
>> > +                               holder->u.qom_object.qom_path);
>> > +                break;
>> > +            default:
>> > +                monitor_printf(mon, "[unknown]");
>> > +                break;
>>
>> This is unreachable.  Please use
>>
>>                default:
>>                    g_assert_not_reached();
>>
>> to make it obvious.
>>
>> > +            }
>> > +        }
>> > +        monitor_printf(mon, "\n");
>> > +
>> >          monitor_printf(mon, "  poll-max-ns=%" PRId64 "\n", value->poll_max_ns);
>> >          monitor_printf(mon, "  poll-grow=%" PRId64 "\n", value->poll_grow);
>> >          monitor_printf(mon, "  poll-shrink=%" PRId64 "\n", value->poll_shrink);
>>
>> Quick test:
>>
>>     $ qemu-system-x86_64 -S -nodefaults -display none -chardev socket,id=chr-qmp,path=test-qmp,server=on,wait=off -object monitor-qmp,id=qmp,chardev=chr-qmp -monitor stdio
>>     QEMU 11.1.50 monitor - type 'help' for more information
>>     (qemu) info iothreads
>>     (qemu) q
>>
>> Hmm, no output?  The QMP monitor should be holding the monitor I/O
>> thread.  To confirm that, I made monitor_complete() print things (patch
>> appended), and got
>>
>>     ### monitor qmp holding iothread mon_iothread
>>     ### monitor compat_monitor0 without iothread
>>
>> What's going on here?
>>
>
> This is expected with the existing query-iothreads behavior.
> mon_iothread is created under the internal object root, while
> qmp_query_iothreads() only walks /objects, so internal IOThreads are
> intentionally not exposed through QMP/HMP.

TIL about the internal object root.  It's the root of a *second* QOM
composition tree.  I have been operating under the assumnption there is
just one.

Your series adds holder tracking for any I/O thread, whether it's in the
externally visible QOM composition tree (the one returned by
object_get_root()) or the internal one (the one returned by
object_get_internal_root()).  Makes sense, because at the place where we
track we don't want to know / make assumptions about which tree it is.

query-iothreads shows only the I/O threads in the externally visible QOM
composition tree.  Makes sense.

Fortunately, you were able to eliminate the special case for monitors.
If it was still there, the query-iothreads code to show it would be
dead.

Back to what got us there.  My attempt to test query-iothreads showing
holders failed, because monitors are not usable for this test.  Please
show me how to test this.  Ideally, a simple command line and a few
simple monitor commands ending with a query-iothreads that shows
holders.

> I prefer to preserve that behavior in this series, since exposing internal
> IOThreads would require additional API decisions, including handling ID
> collisions with user-created IOThreads.

Okay!

> I’ll also change the unreachable default case to
> g_assert_not_reached().
>
> Thanks
> Chen

[...]