On Thu, Nov 21, 2024 at 02:01:45PM +0100, Philippe Mathieu-Daudé wrote:
> On 21/11/24 11:30, Daniel P. Berrangé wrote:
> > On Wed, Nov 20, 2024 at 04:57:01PM -0500, Peter Xu wrote:
> > > Always explicitly create QEMU system containers upfront.
> > >
> > > Root containers will be created when trying to fetch the root object the
> > > 1st time. Machine sub-containers will be created only until machine is
> > > being initialized.
> > >
> > > Signed-off-by: Peter Xu <[email protected]>
> > > ---
> > > hw/core/machine.c | 19 ++++++++++++++++---
> > > qom/object.c | 16 +++++++++++++++-
> > > 2 files changed, 31 insertions(+), 4 deletions(-)
> >
> >
> > > diff --git a/qom/object.c b/qom/object.c
> > > index 214d6eb4c1..810e6f2bd9 100644
> > > --- a/qom/object.c
> > > +++ b/qom/object.c
> > > @@ -1734,12 +1734,26 @@ const char *object_property_get_type(Object *obj,
> > > const char *name, Error **errp
> > > return prop->type;
> > > }
> > > +static Object *object_root_initialize(void)
> > > +{
> > > + Object *root = object_new(TYPE_CONTAINER);
> > > +
> > > + /*
> > > + * Create all QEMU system containers. "machine" and its
> > > sub-containers
> > > + * are only created when machine initializes (qemu_create_machine()).
> > > + */
> > > + container_create(root, "chardevs");
> > > + container_create(root, "objects");
> >
> > This is where I would expect 'backend' to have been created
> > rather than ui/console.c, though you could potentially make
> > a case to create it from the machine function, snice console
> > stuff can't be used outside of the machine context, while
> > chardevs/objects can be used in qemu-img/qemu-nbd, etc
Would it hurt if we do it altogether here even if it won't be used in
qemu-img/qemu-nbd?
IMHO we should either make it simple (assuming empty containers won't hurt
there..), or we should just leave "backend" to ui/ code, so we don't assume
which binary is using the ui code: whoever uses it will create the container.
>
> What about creating "backend" container in qemu_create_machine()?
I remember I started with that but it didn't work. IIRC that's because
machine_initfn() (or somewhere around the init code) requires the
containers to present, hence it's too late even if we create the containers
right after this line:
current_machine =
MACHINE(object_new_with_class(OBJECT_CLASS(machine_class)));
--
Peter Xu