> On Wed, 29 Apr 2026 14:48:36 +0100 > Joshua Lant <[email protected]> wrote: > > > Extend the capability for hiding devices, introduced for virtio-net > > device in: > > > > commit f3a8505656935cde32e28c1c6317f725084da1e0 > > Author: Jens Freimann <[email protected]> > > Date: Tue Oct 29 12:48:55 2019 +0100 > > qdev/qbus: add hidden device support > > > > Currently only endpoint devices can be hidden with a primary device > > and failover (known static configuration). However, looking at future > > composable systems, we see a need for hidden devices which have no > > associated > > bus upon boot. Move the check for hidden devices to before the bus > > search, and if it is hidden ignore the case where the device was > > described on the CLI without the "bus=" field. > > > > This is motivated by a specific use-case: implementing the VCS > > command set, part of the CXL specification (CXL r3.2 Section 7.1.3). In > > this scenario a switch controlled by a Fabric Manager is able to change the > > virtual hierarchy of devices seen by a guest within a fixed physical system > > topology. The connecting bus is not known until runtime when > > a bind command is issued by the Fabric Manager. > > So this comes back to the earlier comment on whether they are completely > hidden from a guest. They aren't quite - we need to be able to > get to them via the MCTP or cxl-switch-cci (in band mailbox) paths. > > Maybe we can solve that later without any compatibility problems > or by cheating and just searching hidden devices. I'm not sure.
Funnily enough I had something like this when I was originally writing this series. I scrapped it becasue I ran into issues. I was storing as a tree the QDicts of hidden busses and devices below them, but then realised I couldnt hotplug into the USP and abandoned it. I was doing this with the aim of eventual support for cascaded switches. There was another issue I cant fully remember... That the searching was easy when it was referring to unrealized devices with QDicts from the CLI, but it became much more difficult to get proper reference to some structures I needed from the devices inside the qdev/qbus code when they were realized... > > > > Signed-off-by: Joshua Lant <[email protected]> > > --- > > system/qdev-monitor.c | 10 ++++++++-- > > 1 file changed, 8 insertions(+), 2 deletions(-) > > > > diff --git a/system/qdev-monitor.c b/system/qdev-monitor.c > > index f2aa400a77..b51dfe0645 100644 > > --- a/system/qdev-monitor.c > > +++ b/system/qdev-monitor.c > > @@ -650,6 +650,7 @@ DeviceState *qdev_device_add_from_qdict(const QDict > > *opts, > > DeviceState *dev; > > BusState *bus = NULL; > > QDict *properties; > > + bool hide_device; > > > > driver = qdict_get_try_str(opts, "driver"); > > if (!driver) { > > @@ -663,6 +664,11 @@ DeviceState *qdev_device_add_from_qdict(const QDict > > *opts, > > return NULL; > > } > > > > + /* Is the device hidden from the guest? > > + * If yes, no need to find a default bus if none given... > > + * Bus could be provided at runtime (i.e. in a switch)*/ > > + hide_device = qdev_should_hide_device(opts, from_json, errp); > > + > > /* find bus */ > > path = qdict_get_try_str(opts, "bus"); > > if (path != NULL) { > > @@ -675,14 +681,14 @@ DeviceState *qdev_device_add_from_qdict(const QDict > > *opts, > > driver, object_get_typename(OBJECT(bus))); > > return NULL; > > } > > - } else if (dc->bus_type != NULL) { > > + } else if (dc->bus_type != NULL && !hide_device) { > > bus = qdev_find_default_bus(dc, errp); > > if (!bus) { > > return NULL; > > } > > } > > > > - if (qdev_should_hide_device(opts, from_json, errp)) { > > + if (hide_device) { > > if (bus && !qbus_is_hotpluggable(bus)) { > > error_setg(errp, "Bus '%s' does not support hotplugging", > > bus->name); >
