All of lore.kernel.org
 help / color / mirror / Atom feed
* x-remote-object breaks QOM introspection
@ 2026-11-17 14:58 Markus Armbruster
  2026-11-18 15:27 ` Jag Raman
  0 siblings, 1 reply; 7+ messages in thread
From: Markus Armbruster @ 2026-11-17 14:58 UTC (permalink / raw)
  To: Elena Ufimtseva, Jagannathan Raman
  Cc: qemu-devel, Paolo Bonzini, Daniel P. Berrangé,
	Philippe Mathieu-Daudé, Zhao Liu

Watch this:

    $ qemu-system-x86_64 -S -nodefaults -display none -qmp stdio
    {"QMP": {"version": {"qemu": {"micro": 90, "minor": 1, "major": 11}, "package": "v11.2.0-rc0-3-g38171f59ccb-dirty"}, "capabilities": ["oob"]}}
    {"execute": "qmp_capabilities", "arguments": {"enable": ["oob"]}}
    {"return": {}}
    {"execute": "qom-list-properties", "arguments": {"typename": "x-remote-object"}}
--> qemu-system-x86_64: (null) is not a PCI device
    {"return": [{"name": "type", "type": "string", "qapi-type": "str"}, {"name": "devid", "type": "string", "qapi-type": "str"}, {"name": "fd", "type": "string", "qapi-type": "str"}]}

The error message is suspicious.  Origin:

    static void remote_object_machine_done(Notifier *notifier, void *data)
    {
        RemoteObject *o = container_of(notifier, RemoteObject, machine_done);
        DeviceState *dev = NULL;
        QIOChannel *ioc = NULL;
        Coroutine *co = NULL;
        RemoteCommDev *comdev = NULL;
        Error *err = NULL;

        dev = qdev_find_recursive(sysbus_get_default(), o->devid);
        if (!dev || !object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
            error_report("%s is not a PCI device", o->devid);
            return;
        }

o->devid is null, dev is null.

Glibc's %s shows null pointer as "(null)".  There are libcs that simply
crash.

Stack backtrace:

    #0  remote_object_machine_done (notifier=0x5555587d17e8, data=0x0)
        at ../hw/remote/remote-obj.c:93
    #1  0x000055555578700e in qemu_add_machine_init_done_notifier
        (notify=0x5555587d17e8) at ../hw/core/machine.c:1721
    #2  0x00005555556f26dd in remote_object_init (obj=0x5555587d17c0)
        at ../hw/remote/remote-obj.c:149
    #3  0x0000555555cbda2d in object_init_with_type
        (obj=0x5555587d17c0, ti=0x5555577c02d0) at ../qom/object.c:430
    #4  0x0000555555cbde2c in object_initialize_with_type
        (obj=0x5555587d17c0, size=136, type=0x5555577c02d0) at ../qom/object.c:512
    #5  0x0000555555cbe6fa in object_new_with_type (type=0x5555577c02d0)
        at ../qom/object.c:710
    #6  0x0000555555cbe784 in object_new
        (typename=0x555558356a10 "x-remote-object") at ../qom/object.c:726
    #7  0x0000555555e2d6be in qmp_qom_list_properties
        (typename=0x555558356a10 "x-remote-object", errp=0x7fffffffd888)
        at ../qom/qom-qmp-cmds.c:290
    [...]

I believe using qemu_add_machine_init_done_notifier() is wrong here.
Could we use UserCreatableClass method complete()?

Aside: we should probably check all uses of
qemu_add_machine_init_done_notifier().


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: x-remote-object breaks QOM introspection
  2026-11-17 14:58 x-remote-object breaks QOM introspection Markus Armbruster
@ 2026-11-18 15:27 ` Jag Raman
  2026-11-19  6:07   ` Markus Armbruster
  0 siblings, 1 reply; 7+ messages in thread
From: Jag Raman @ 2026-11-18 15:27 UTC (permalink / raw)
  To: Markus Armbruster
  Cc: Elena Ufimtseva, qemu-devel@nongnu.org, Paolo Bonzini,
	Daniel P. Berrangé, Philippe Mathieu-Daudé, Zhao Liu



> On Nov 17, 2026, at 9:58 AM, Markus Armbruster <armbru@pond.sub.org> wrote:
> 
> This Message Is From a New Sender
> You have not previously corresponded with the sender of this message.
> Report Suspicious
> Watch this:
> 
> $ qemu-system-x86_64 -S -nodefaults -display none -qmp stdio
> {"QMP": {"version": {"qemu": {"micro": 90, "minor": 1, "major": 11}, "package": "v11.2.0-rc0-3-g38171f59ccb-dirty"}, "capabilities": ["oob"]}}
> {"execute": "qmp_capabilities", "arguments": {"enable": ["oob"]}}
> {"return": {}}
> {"execute": "qom-list-properties", "arguments": {"typename": "x-remote-object"}}
> --> qemu-system-x86_64: (null) is not a PCI device
> {"return": [{"name": "type", "type": "string", "qapi-type": "str"}, {"name": "devid", "type": "string", "qapi-type": "str"}, {"name": "fd", "type": "string", "qapi-type": "str"}]}
> 
> The error message is suspicious. Origin:
> 
> static void remote_object_machine_done(Notifier *notifier, void *data)
> {
> RemoteObject *o = container_of(notifier, RemoteObject, machine_done);
> DeviceState *dev = NULL;
> QIOChannel *ioc = NULL;
> Coroutine *co = NULL;
> RemoteCommDev *comdev = NULL;
> Error *err = NULL;
> 
> dev = qdev_find_recursive(sysbus_get_default(), o->devid);
> if (!dev || !object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
> error_report("%s is not a PCI device", o->devid);
> return;
> }
> 
> o->devid is null, dev is null.
> 
> Glibc's %s shows null pointer as "(null)". There are libcs that simply
> crash.
> 
> Stack backtrace:
> 
> #0 remote_object_machine_done (notifier=0x5555587d17e8, data=0x0)
> at ../hw/remote/remote-obj.c:93
> #1 0x000055555578700e in qemu_add_machine_init_done_notifier
> (notify=0x5555587d17e8) at ../hw/core/machine.c:1721
> #2 0x00005555556f26dd in remote_object_init (obj=0x5555587d17c0)
> at ../hw/remote/remote-obj.c:149
> #3 0x0000555555cbda2d in object_init_with_type
> (obj=0x5555587d17c0, ti=0x5555577c02d0) at ../qom/object.c:430
> #4 0x0000555555cbde2c in object_initialize_with_type
> (obj=0x5555587d17c0, size=136, type=0x5555577c02d0) at ../qom/object.c:512
> #5 0x0000555555cbe6fa in object_new_with_type (type=0x5555577c02d0)
> at ../qom/object.c:710
> #6 0x0000555555cbe784 in object_new
> (typename=0x555558356a10 "x-remote-object") at ../qom/object.c:726
> #7 0x0000555555e2d6be in qmp_qom_list_properties
> (typename=0x555558356a10 "x-remote-object", errp=0x7fffffffd888)
> at ../qom/qom-qmp-cmds.c:290
> [...]
> 
> I believe using qemu_add_machine_init_done_notifier() is wrong here.
> Could we use UserCreatableClass method complete()?
> 

Hi Markus,

The remote_object_machine_done() notifier exists so qdev_find_recursive()
has a chance to find o->devid. We could setup this notifier UserCreatableClass
method complete() if you prefer.

I think remote_object_machine_done() should return early if o->devid is
NULL. Would you like me to make this change?

Thank you,
Jag

> 
> Aside: we should probably check all uses of
> qemu_add_machine_init_done_notifier().
> 


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: x-remote-object breaks QOM introspection
  2026-11-18 15:27 ` Jag Raman
@ 2026-11-19  6:07   ` Markus Armbruster
  2026-11-19 16:13     ` Jag Raman
  0 siblings, 1 reply; 7+ messages in thread
From: Markus Armbruster @ 2026-11-19  6:07 UTC (permalink / raw)
  To: Jag Raman
  Cc: Elena Ufimtseva, qemu-devel@nongnu.org, Paolo Bonzini,
	Daniel P. Berrangé, Philippe Mathieu-Daudé, Zhao Liu

Jag Raman <jag.raman@oracle.com> writes:

>> On Nov 17, 2026, at 9:58 AM, Markus Armbruster <armbru@pond.sub.org> wrote:
>> 
>> Watch this:
>> 
>>     $ qemu-system-x86_64 -S -nodefaults -display none -qmp stdio
>>     {"QMP": {"version": {"qemu": {"micro": 90, "minor": 1, "major": 11}, "package": "v11.2.0-rc0-3-g38171f59ccb-dirty"}, "capabilities": ["oob"]}}
>>     {"execute": "qmp_capabilities", "arguments": {"enable": ["oob"]}}
>>     {"return": {}}
>>     {"execute": "qom-list-properties", "arguments": {"typename": "x-remote-object"}}
>> --> qemu-system-x86_64: (null) is not a PCI device
>>     {"return": [{"name": "type", "type": "string", "qapi-type": "str"}, {"name": "devid", "type": "string", "qapi-type": "str"}, {"name": "fd", "type": "string", "qapi-type": "str"}]}
>> 
>> The error message is suspicious.  Origin:
>> 
>>     static void remote_object_machine_done(Notifier *notifier, void *data)
>>     {
>>         RemoteObject *o = container_of(notifier, RemoteObject, machine_done);
>>         DeviceState *dev = NULL;
>>         QIOChannel *ioc = NULL;
>>         Coroutine *co = NULL;
>>         RemoteCommDev *comdev = NULL;
>>         Error *err = NULL;
>> 
>>         dev = qdev_find_recursive(sysbus_get_default(), o->devid);
>>         if (!dev || !object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
>>             error_report("%s is not a PCI device", o->devid);
>>             return;
>>         }
>> 
>> o->devid is null, dev is null.
>> 
>> Glibc's %s shows null pointer as "(null)".  There are libcs that simply
>> crash.
>> 
>> Stack backtrace:
>> 
>>     #0  remote_object_machine_done (notifier=0x5555587d17e8, data=0x0)
>>         at ../hw/remote/remote-obj.c:93
>>     #1  0x000055555578700e in qemu_add_machine_init_done_notifier
>>         (notify=0x5555587d17e8) at ../hw/core/machine.c:1721
>>     #2  0x00005555556f26dd in remote_object_init (obj=0x5555587d17c0)
>>         at ../hw/remote/remote-obj.c:149
>>     #3  0x0000555555cbda2d in object_init_with_type
>>         (obj=0x5555587d17c0, ti=0x5555577c02d0) at ../qom/object.c:430
>>     #4  0x0000555555cbde2c in object_initialize_with_type
>>         (obj=0x5555587d17c0, size=136, type=0x5555577c02d0) at ../qom/object.c:512
>>     #5  0x0000555555cbe6fa in object_new_with_type (type=0x5555577c02d0)
>>         at ../qom/object.c:710
>>     #6  0x0000555555cbe784 in object_new
>>         (typename=0x555558356a10 "x-remote-object") at ../qom/object.c:726
>>     #7  0x0000555555e2d6be in qmp_qom_list_properties
>>         (typename=0x555558356a10 "x-remote-object", errp=0x7fffffffd888)
>>         at ../qom/qom-qmp-cmds.c:290
>>     [...]
>> 
>> I believe using qemu_add_machine_init_done_notifier() is wrong here.
>> Could we use UserCreatableClass method complete()?
>
> Hi Markus,
>
> The remote_object_machine_done() notifier exists so qdev_find_recursive()
> has a chance to find o->devid. We could setup this notifier UserCreatableClass
> method complete() if you prefer.

I guess that unbreaks QOM introspection, because complete() doesn't run
there.

Can you explain why you need to defer code to machine init done?

> I think remote_object_machine_done() should return early if o->devid is
> NULL. Would you like me to make this change?

Should devid be a QOM link instead?

> Thank you,
> Jag
>
>> 
>> Aside: we should probably check all uses of
>> qemu_add_machine_init_done_notifier().


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: x-remote-object breaks QOM introspection
  2026-11-19  6:07   ` Markus Armbruster
@ 2026-11-19 16:13     ` Jag Raman
  2026-11-19 16:25       ` Daniel P. Berrangé
  0 siblings, 1 reply; 7+ messages in thread
From: Jag Raman @ 2026-11-19 16:13 UTC (permalink / raw)
  To: Markus Armbruster
  Cc: Elena Ufimtseva, qemu-devel@nongnu.org, Paolo Bonzini,
	Daniel P. Berrangé, Philippe Mathieu-Daudé, Zhao Liu



> On Nov 19, 2026, at 1:07 AM, Markus Armbruster <armbru@pond.sub.org> wrote:
> 
> This Message Is From a New Sender
> You have not previously corresponded with the sender of this message.
> Report Suspicious
> Jag Raman <jag.raman@oracle.com> writes:
> 
> >> On Nov 17, 2026, at 9:58 AM, Markus Armbruster <armbru@pond.sub.org> wrote:
> >> 
> >> Watch this:
> >> 
> >> $ qemu-system-x86_64 -S -nodefaults -display none -qmp stdio
> >> {"QMP": {"version": {"qemu": {"micro": 90, "minor": 1, "major": 11}, "package": "v11.2.0-rc0-3-g38171f59ccb-dirty"}, "capabilities": ["oob"]}}
> >> {"execute": "qmp_capabilities", "arguments": {"enable": ["oob"]}}
> >> {"return": {}}
> >> {"execute": "qom-list-properties", "arguments": {"typename": "x-remote-object"}}
> >> --> qemu-system-x86_64: (null) is not a PCI device
> >> {"return": [{"name": "type", "type": "string", "qapi-type": "str"}, {"name": "devid", "type": "string", "qapi-type": "str"}, {"name": "fd", "type": "string", "qapi-type": "str"}]}
> >> 
> >> The error message is suspicious. Origin:
> >> 
> >> static void remote_object_machine_done(Notifier *notifier, void *data)
> >> {
> >> RemoteObject *o = container_of(notifier, RemoteObject, machine_done);
> >> DeviceState *dev = NULL;
> >> QIOChannel *ioc = NULL;
> >> Coroutine *co = NULL;
> >> RemoteCommDev *comdev = NULL;
> >> Error *err = NULL;
> >> 
> >> dev = qdev_find_recursive(sysbus_get_default(), o->devid);
> >> if (!dev || !object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
> >> error_report("%s is not a PCI device", o->devid);
> >> return;
> >> }
> >> 
> >> o->devid is null, dev is null.
> >> 
> >> Glibc's %s shows null pointer as "(null)". There are libcs that simply
> >> crash.
> >> 
> >> Stack backtrace:
> >> 
> >> #0 remote_object_machine_done (notifier=0x5555587d17e8, data=0x0)
> >> at ../hw/remote/remote-obj.c:93
> >> #1 0x000055555578700e in qemu_add_machine_init_done_notifier
> >> (notify=0x5555587d17e8) at ../hw/core/machine.c:1721
> >> #2 0x00005555556f26dd in remote_object_init (obj=0x5555587d17c0)
> >> at ../hw/remote/remote-obj.c:149
> >> #3 0x0000555555cbda2d in object_init_with_type
> >> (obj=0x5555587d17c0, ti=0x5555577c02d0) at ../qom/object.c:430
> >> #4 0x0000555555cbde2c in object_initialize_with_type
> >> (obj=0x5555587d17c0, size=136, type=0x5555577c02d0) at ../qom/object.c:512
> >> #5 0x0000555555cbe6fa in object_new_with_type (type=0x5555577c02d0)
> >> at ../qom/object.c:710
> >> #6 0x0000555555cbe784 in object_new
> >> (typename=0x555558356a10 "x-remote-object") at ../qom/object.c:726
> >> #7 0x0000555555e2d6be in qmp_qom_list_properties
> >> (typename=0x555558356a10 "x-remote-object", errp=0x7fffffffd888)
> >> at ../qom/qom-qmp-cmds.c:290
> >> [...]
> >> 
> >> I believe using qemu_add_machine_init_done_notifier() is wrong here.
> >> Could we use UserCreatableClass method complete()?
> >
> > Hi Markus,
> >
> > The remote_object_machine_done() notifier exists so qdev_find_recursive()
> > has a chance to find o->devid. We could setup this notifier UserCreatableClass
> > method complete() if you prefer.
> 
> I guess that unbreaks QOM introspection, because complete() doesn't run
> there.
> 
> Can you explain why you need to defer code to machine init done?

When x-remote-object is instantiated, the PCI device identified by o->devid may not
yet exist because it appears later in QEMU’s command-line options. As a result,
qdev_find_recursive() may fail to find it. Deferring the lookup until machine init done
ensures that all devices have been instantiated and can be found using qdev_find_recursive().

> 
> > I think remote_object_machine_done() should return early if o->devid is
> > NULL. Would you like me to make this change?
> 
> Should devid be a QOM link instead?

QOM link sounds like a good idea. However, it’s unlikely to remove the wait for machine
init done from what I can tell.

> 
> > Thank you,
> > Jag
> >
> >> 
> >> Aside: we should probably check all uses of
> >> qemu_add_machine_init_done_notifier().
> 


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: x-remote-object breaks QOM introspection
  2026-11-19 16:13     ` Jag Raman
@ 2026-11-19 16:25       ` Daniel P. Berrangé
  2026-11-20  5:21         ` Markus Armbruster
  0 siblings, 1 reply; 7+ messages in thread
From: Daniel P. Berrangé @ 2026-11-19 16:25 UTC (permalink / raw)
  To: Jag Raman
  Cc: Markus Armbruster, Elena Ufimtseva, qemu-devel@nongnu.org,
	Paolo Bonzini, Philippe Mathieu-Daudé, Zhao Liu

On Thu, Nov 19, 2026 at 04:13:31PM +0000, Jag Raman wrote:
> 
> 
> > On Nov 19, 2026, at 1:07 AM, Markus Armbruster <armbru@pond.sub.org> wrote:
> > 
> > This Message Is From a New Sender
> > You have not previously corresponded with the sender of this message.
> > Report Suspicious
> > Jag Raman <jag.raman@oracle.com> writes:
> > 
> > >> On Nov 17, 2026, at 9:58 AM, Markus Armbruster <armbru@pond.sub.org> wrote:
> > >> 
> > >> Watch this:
> > >> 
> > >> $ qemu-system-x86_64 -S -nodefaults -display none -qmp stdio
> > >> {"QMP": {"version": {"qemu": {"micro": 90, "minor": 1, "major": 11}, "package": "v11.2.0-rc0-3-g38171f59ccb-dirty"}, "capabilities": ["oob"]}}
> > >> {"execute": "qmp_capabilities", "arguments": {"enable": ["oob"]}}
> > >> {"return": {}}
> > >> {"execute": "qom-list-properties", "arguments": {"typename": "x-remote-object"}}
> > >> --> qemu-system-x86_64: (null) is not a PCI device
> > >> {"return": [{"name": "type", "type": "string", "qapi-type": "str"}, {"name": "devid", "type": "string", "qapi-type": "str"}, {"name": "fd", "type": "string", "qapi-type": "str"}]}
> > >> 
> > >> The error message is suspicious. Origin:
> > >> 
> > >> static void remote_object_machine_done(Notifier *notifier, void *data)
> > >> {
> > >> RemoteObject *o = container_of(notifier, RemoteObject, machine_done);
> > >> DeviceState *dev = NULL;
> > >> QIOChannel *ioc = NULL;
> > >> Coroutine *co = NULL;
> > >> RemoteCommDev *comdev = NULL;
> > >> Error *err = NULL;
> > >> 
> > >> dev = qdev_find_recursive(sysbus_get_default(), o->devid);
> > >> if (!dev || !object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
> > >> error_report("%s is not a PCI device", o->devid);
> > >> return;
> > >> }
> > >> 
> > >> o->devid is null, dev is null.
> > >> 
> > >> Glibc's %s shows null pointer as "(null)". There are libcs that simply
> > >> crash.
> > >> 
> > >> Stack backtrace:
> > >> 
> > >> #0 remote_object_machine_done (notifier=0x5555587d17e8, data=0x0)
> > >> at ../hw/remote/remote-obj.c:93
> > >> #1 0x000055555578700e in qemu_add_machine_init_done_notifier
> > >> (notify=0x5555587d17e8) at ../hw/core/machine.c:1721
> > >> #2 0x00005555556f26dd in remote_object_init (obj=0x5555587d17c0)
> > >> at ../hw/remote/remote-obj.c:149
> > >> #3 0x0000555555cbda2d in object_init_with_type
> > >> (obj=0x5555587d17c0, ti=0x5555577c02d0) at ../qom/object.c:430
> > >> #4 0x0000555555cbde2c in object_initialize_with_type
> > >> (obj=0x5555587d17c0, size=136, type=0x5555577c02d0) at ../qom/object.c:512
> > >> #5 0x0000555555cbe6fa in object_new_with_type (type=0x5555577c02d0)
> > >> at ../qom/object.c:710
> > >> #6 0x0000555555cbe784 in object_new
> > >> (typename=0x555558356a10 "x-remote-object") at ../qom/object.c:726
> > >> #7 0x0000555555e2d6be in qmp_qom_list_properties
> > >> (typename=0x555558356a10 "x-remote-object", errp=0x7fffffffd888)
> > >> at ../qom/qom-qmp-cmds.c:290
> > >> [...]
> > >> 
> > >> I believe using qemu_add_machine_init_done_notifier() is wrong here.
> > >> Could we use UserCreatableClass method complete()?
> > >
> > > Hi Markus,
> > >
> > > The remote_object_machine_done() notifier exists so qdev_find_recursive()
> > > has a chance to find o->devid. We could setup this notifier UserCreatableClass
> > > method complete() if you prefer.
> > 
> > I guess that unbreaks QOM introspection, because complete() doesn't run
> > there.
> > 
> > Can you explain why you need to defer code to machine init done?
> 
> When x-remote-object is instantiated, the PCI device identified by o->devid may not
> yet exist because it appears later in QEMU’s command-line options. As a result,
> qdev_find_recursive() may fail to find it. Deferring the lookup until machine init done
> ensures that all devices have been instantiated and can be found using qdev_find_recursive().

We've already got hacks to process different -object classes at
different phases of command line processing. It sounds like
x-remote-object needs to be set to be processed after -device.

ie we need a new object_create_very_late() to be run after
qemu_create_cli_devices()

With regards,
Daniel
-- 
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|



^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: x-remote-object breaks QOM introspection
  2026-11-19 16:25       ` Daniel P. Berrangé
@ 2026-11-20  5:21         ` Markus Armbruster
  2026-11-20  8:53           ` Daniel P. Berrangé
  0 siblings, 1 reply; 7+ messages in thread
From: Markus Armbruster @ 2026-11-20  5:21 UTC (permalink / raw)
  To: Daniel P. Berrangé
  Cc: Jag Raman, Elena Ufimtseva, qemu-devel@nongnu.org, Paolo Bonzini,
	Philippe Mathieu-Daudé, Zhao Liu

Daniel P. Berrangé <berrange@redhat.com> writes:

> On Thu, Nov 19, 2026 at 04:13:31PM +0000, Jag Raman wrote:
>> 
>> 
>> > On Nov 19, 2026, at 1:07 AM, Markus Armbruster <armbru@pond.sub.org> wrote:
>> > 
>> > This Message Is From a New Sender
>> > You have not previously corresponded with the sender of this message.
>> > Report Suspicious
>> > Jag Raman <jag.raman@oracle.com> writes:
>> > 
>> > >> On Nov 17, 2026, at 9:58 AM, Markus Armbruster <armbru@pond.sub.org> wrote:
>> > >> 
>> > >> Watch this:
>> > >> 
>> > >> $ qemu-system-x86_64 -S -nodefaults -display none -qmp stdio
>> > >> {"QMP": {"version": {"qemu": {"micro": 90, "minor": 1, "major": 11}, "package": "v11.2.0-rc0-3-g38171f59ccb-dirty"}, "capabilities": ["oob"]}}
>> > >> {"execute": "qmp_capabilities", "arguments": {"enable": ["oob"]}}
>> > >> {"return": {}}
>> > >> {"execute": "qom-list-properties", "arguments": {"typename": "x-remote-object"}}
>> > >> --> qemu-system-x86_64: (null) is not a PCI device
>> > >> {"return": [{"name": "type", "type": "string", "qapi-type": "str"}, {"name": "devid", "type": "string", "qapi-type": "str"}, {"name": "fd", "type": "string", "qapi-type": "str"}]}
>> > >> 
>> > >> The error message is suspicious. Origin:
>> > >> 
>> > >> static void remote_object_machine_done(Notifier *notifier, void *data)
>> > >> {
>> > >> RemoteObject *o = container_of(notifier, RemoteObject, machine_done);
>> > >> DeviceState *dev = NULL;
>> > >> QIOChannel *ioc = NULL;
>> > >> Coroutine *co = NULL;
>> > >> RemoteCommDev *comdev = NULL;
>> > >> Error *err = NULL;
>> > >> 
>> > >> dev = qdev_find_recursive(sysbus_get_default(), o->devid);
>> > >> if (!dev || !object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
>> > >> error_report("%s is not a PCI device", o->devid);
>> > >> return;
>> > >> }
>> > >> 
>> > >> o->devid is null, dev is null.
>> > >> 
>> > >> Glibc's %s shows null pointer as "(null)". There are libcs that simply
>> > >> crash.
>> > >> 
>> > >> Stack backtrace:
>> > >> 
>> > >> #0 remote_object_machine_done (notifier=0x5555587d17e8, data=0x0)
>> > >> at ../hw/remote/remote-obj.c:93
>> > >> #1 0x000055555578700e in qemu_add_machine_init_done_notifier
>> > >> (notify=0x5555587d17e8) at ../hw/core/machine.c:1721
>> > >> #2 0x00005555556f26dd in remote_object_init (obj=0x5555587d17c0)
>> > >> at ../hw/remote/remote-obj.c:149
>> > >> #3 0x0000555555cbda2d in object_init_with_type
>> > >> (obj=0x5555587d17c0, ti=0x5555577c02d0) at ../qom/object.c:430
>> > >> #4 0x0000555555cbde2c in object_initialize_with_type
>> > >> (obj=0x5555587d17c0, size=136, type=0x5555577c02d0) at ../qom/object.c:512
>> > >> #5 0x0000555555cbe6fa in object_new_with_type (type=0x5555577c02d0)
>> > >> at ../qom/object.c:710
>> > >> #6 0x0000555555cbe784 in object_new
>> > >> (typename=0x555558356a10 "x-remote-object") at ../qom/object.c:726
>> > >> #7 0x0000555555e2d6be in qmp_qom_list_properties
>> > >> (typename=0x555558356a10 "x-remote-object", errp=0x7fffffffd888)
>> > >> at ../qom/qom-qmp-cmds.c:290
>> > >> [...]
>> > >> 
>> > >> I believe using qemu_add_machine_init_done_notifier() is wrong here.
>> > >> Could we use UserCreatableClass method complete()?
>> > >
>> > > Hi Markus,
>> > >
>> > > The remote_object_machine_done() notifier exists so qdev_find_recursive()
>> > > has a chance to find o->devid. We could setup this notifier UserCreatableClass
>> > > method complete() if you prefer.
>> > 
>> > I guess that unbreaks QOM introspection, because complete() doesn't run
>> > there.
>> > 
>> > Can you explain why you need to defer code to machine init done?
>> 
>> When x-remote-object is instantiated, the PCI device identified by o->devid may not
>> yet exist because it appears later in QEMU’s command-line options. As a result,
>> qdev_find_recursive() may fail to find it. Deferring the lookup until machine init done
>> ensures that all devices have been instantiated and can be found using qdev_find_recursive().
>
> We've already got hacks to process different -object classes at
> different phases of command line processing. It sounds like
> x-remote-object needs to be set to be processed after -device.
>
> ie we need a new object_create_very_late() to be run after
> qemu_create_cli_devices()

Many devices interact with supporting objects.  We have a common
frontend / backend pattern:

* The device (frontend) has a property identifying the object (backend).
  The property setter resolves the property value to an object of the
  required type, and fails if it can't.  This also fails device
  creation.

  CLI works, because we execute -object before -device.

  Monitor works as long as users object_add before device_add.

* Ideally, the device can work with any object of the required kind.
  When it can't, which isn't rare, its .realize() checks the object is
  suitable, and fails if it isn't.  This also fails device creation.

Why isn't this workable here?


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: x-remote-object breaks QOM introspection
  2026-11-20  5:21         ` Markus Armbruster
@ 2026-11-20  8:53           ` Daniel P. Berrangé
  0 siblings, 0 replies; 7+ messages in thread
From: Daniel P. Berrangé @ 2026-11-20  8:53 UTC (permalink / raw)
  To: Markus Armbruster
  Cc: Jag Raman, Elena Ufimtseva, qemu-devel@nongnu.org, Paolo Bonzini,
	Philippe Mathieu-Daudé, Zhao Liu

On Fri, Nov 20, 2026 at 06:21:57AM +0100, Markus Armbruster wrote:
> Daniel P. Berrangé <berrange@redhat.com> writes:
> 
> > On Thu, Nov 19, 2026 at 04:13:31PM +0000, Jag Raman wrote:
> >> 
> >> 
> >> > On Nov 19, 2026, at 1:07 AM, Markus Armbruster <armbru@pond.sub.org> wrote:
> >> > 
> >> > This Message Is From a New Sender
> >> > You have not previously corresponded with the sender of this message.
> >> > Report Suspicious
> >> > Jag Raman <jag.raman@oracle.com> writes:
> >> > 
> >> > >> On Nov 17, 2026, at 9:58 AM, Markus Armbruster <armbru@pond.sub.org> wrote:
> >> > >> 
> >> > >> Watch this:
> >> > >> 
> >> > >> $ qemu-system-x86_64 -S -nodefaults -display none -qmp stdio
> >> > >> {"QMP": {"version": {"qemu": {"micro": 90, "minor": 1, "major": 11}, "package": "v11.2.0-rc0-3-g38171f59ccb-dirty"}, "capabilities": ["oob"]}}
> >> > >> {"execute": "qmp_capabilities", "arguments": {"enable": ["oob"]}}
> >> > >> {"return": {}}
> >> > >> {"execute": "qom-list-properties", "arguments": {"typename": "x-remote-object"}}
> >> > >> --> qemu-system-x86_64: (null) is not a PCI device
> >> > >> {"return": [{"name": "type", "type": "string", "qapi-type": "str"}, {"name": "devid", "type": "string", "qapi-type": "str"}, {"name": "fd", "type": "string", "qapi-type": "str"}]}
> >> > >> 
> >> > >> The error message is suspicious. Origin:
> >> > >> 
> >> > >> static void remote_object_machine_done(Notifier *notifier, void *data)
> >> > >> {
> >> > >> RemoteObject *o = container_of(notifier, RemoteObject, machine_done);
> >> > >> DeviceState *dev = NULL;
> >> > >> QIOChannel *ioc = NULL;
> >> > >> Coroutine *co = NULL;
> >> > >> RemoteCommDev *comdev = NULL;
> >> > >> Error *err = NULL;
> >> > >> 
> >> > >> dev = qdev_find_recursive(sysbus_get_default(), o->devid);
> >> > >> if (!dev || !object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
> >> > >> error_report("%s is not a PCI device", o->devid);
> >> > >> return;
> >> > >> }
> >> > >> 
> >> > >> o->devid is null, dev is null.
> >> > >> 
> >> > >> Glibc's %s shows null pointer as "(null)". There are libcs that simply
> >> > >> crash.
> >> > >> 
> >> > >> Stack backtrace:
> >> > >> 
> >> > >> #0 remote_object_machine_done (notifier=0x5555587d17e8, data=0x0)
> >> > >> at ../hw/remote/remote-obj.c:93
> >> > >> #1 0x000055555578700e in qemu_add_machine_init_done_notifier
> >> > >> (notify=0x5555587d17e8) at ../hw/core/machine.c:1721
> >> > >> #2 0x00005555556f26dd in remote_object_init (obj=0x5555587d17c0)
> >> > >> at ../hw/remote/remote-obj.c:149
> >> > >> #3 0x0000555555cbda2d in object_init_with_type
> >> > >> (obj=0x5555587d17c0, ti=0x5555577c02d0) at ../qom/object.c:430
> >> > >> #4 0x0000555555cbde2c in object_initialize_with_type
> >> > >> (obj=0x5555587d17c0, size=136, type=0x5555577c02d0) at ../qom/object.c:512
> >> > >> #5 0x0000555555cbe6fa in object_new_with_type (type=0x5555577c02d0)
> >> > >> at ../qom/object.c:710
> >> > >> #6 0x0000555555cbe784 in object_new
> >> > >> (typename=0x555558356a10 "x-remote-object") at ../qom/object.c:726
> >> > >> #7 0x0000555555e2d6be in qmp_qom_list_properties
> >> > >> (typename=0x555558356a10 "x-remote-object", errp=0x7fffffffd888)
> >> > >> at ../qom/qom-qmp-cmds.c:290
> >> > >> [...]
> >> > >> 
> >> > >> I believe using qemu_add_machine_init_done_notifier() is wrong here.
> >> > >> Could we use UserCreatableClass method complete()?
> >> > >
> >> > > Hi Markus,
> >> > >
> >> > > The remote_object_machine_done() notifier exists so qdev_find_recursive()
> >> > > has a chance to find o->devid. We could setup this notifier UserCreatableClass
> >> > > method complete() if you prefer.
> >> > 
> >> > I guess that unbreaks QOM introspection, because complete() doesn't run
> >> > there.
> >> > 
> >> > Can you explain why you need to defer code to machine init done?
> >> 
> >> When x-remote-object is instantiated, the PCI device identified by o->devid may not
> >> yet exist because it appears later in QEMU’s command-line options. As a result,
> >> qdev_find_recursive() may fail to find it. Deferring the lookup until machine init done
> >> ensures that all devices have been instantiated and can be found using qdev_find_recursive().
> >
> > We've already got hacks to process different -object classes at
> > different phases of command line processing. It sounds like
> > x-remote-object needs to be set to be processed after -device.
> >
> > ie we need a new object_create_very_late() to be run after
> > qemu_create_cli_devices()
> 
> Many devices interact with supporting objects.  We have a common
> frontend / backend pattern:
> 
> * The device (frontend) has a property identifying the object (backend).
>   The property setter resolves the property value to an object of the
>   required type, and fails if it can't.  This also fails device
>   creation.
> 
>   CLI works, because we execute -object before -device.
> 
>   Monitor works as long as users object_add before device_add.
> 
> * Ideally, the device can work with any object of the required kind.
>   When it can't, which isn't rare, its .realize() checks the object is
>   suitable, and fails if it isn't.  This also fails device creation.
> 
> Why isn't this workable here?

The relationship between the object and device is reversed from
the normal direction in the case of x-remote-object.

Normally an object provides some kind of backend service that
is consumed by the device.

In this case the device can be any PCI endpoint and x-remote-object
is consuming the device, in order to export it over the network
to another "client" QEMU process which uses the x-pci-proxy-dev.


So in the monitor you would use object_add x-remote-object after
device_add <whatever>, and likewise the CLI should create
"x-remote-object" after processing all -device.

With regards,
Daniel
-- 
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|



^ permalink raw reply	[flat|nested] 7+ messages in thread

end of thread, other threads:[~2026-11-20  8:53 UTC | newest]

Thread overview: 7+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-17 14:58 x-remote-object breaks QOM introspection Markus Armbruster
2026-11-18 15:27 ` Jag Raman
2026-11-19  6:07   ` Markus Armbruster
2026-11-19 16:13     ` Jag Raman
2026-11-19 16:25       ` Daniel P. Berrangé
2026-11-20  5:21         ` Markus Armbruster
2026-11-20  8:53           ` Daniel P. Berrangé

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.