Re: [Qemu-devel] [PATCH for-3.2 v4 16/28] hw: apply machine compat properties without touching globals

[Igor Mammedov <imammedo@redhat.com>]

On Tue, 27 Nov 2018 11:35:27 -0200
Eduardo Habkost <ehabkost@redhat.com> wrote:

> On Tue, Nov 27, 2018 at 05:10:05PM +0400, Marc-André Lureau wrote:
> > On Tue, Nov 27, 2018 at 4:57 PM Eduardo Habkost <ehabkost@redhat.com> wrote:  
> > >
> > > On Tue, Nov 27, 2018 at 01:27:49PM +0400, Marc-André Lureau wrote:  
> > > > Similarly to accel properties, move compat properties out of globals
> > > > registration, and apply the machine compat properties during
> > > > device_post_init().
> > > >
> > > > Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>  
> > > [...]  
> > > > diff --git a/hw/core/qdev.c b/hw/core/qdev.c
> > > > index 7066d28271..3b31b2c025 100644
> > > > --- a/hw/core/qdev.c
> > > > +++ b/hw/core/qdev.c
> > > > @@ -971,17 +971,26 @@ static void device_initfn(Object *obj)
> > > >  }
> > > >
> > > >  static const GPtrArray *ac_compat_props;
> > > > +static const GPtrArray *mc_compat_props;
why you didn't use just 'compat_props' for both?
(it would be cleaner have single registry for compat
properties, and the place that takes care of registration
will take care of necessary ordering)

> > > >
> > > >  void accel_register_compat_props(const GPtrArray *props)
> > > >  {
> > > >      ac_compat_props = props;
> > > >  }
> > > >
> > > > +void machine_register_compat_props(const GPtrArray *props)
> > > > +{
> > > > +    mc_compat_props = props;
> > > > +}
> > > > +
> > > >  static void device_post_init(Object *obj)
> > > >  {
> > > >      if (ac_compat_props) {
> > > >          object_apply_global_props(obj, ac_compat_props, &error_abort);
> > > >      }  
> > >
> > > Why not just use MACHINE(qdev_get_machine())->accel->compat_props
> > > directly?
> > >  
> > > > +    if (mc_compat_props) {
> > > > +        object_apply_global_props(obj, mc_compat_props, &error_abort);
> > > > +    }  
> > >
> > > Why not just use MACHINE(qdev_get_machine())->compat_props
> > > directly?  
> > 
> > This was the approach in v3, but Igor didn't quite like referencing
> > machine in qdev:
> > https://lists.nongnu.org/archive/html/qemu-devel/2018-11/msg04774.html  
> 
> I disagree with Igor, here.  Core qdev code already have multiple
> references to machine, I don't see any problem with that.
(There are only 3 calls to qdev_get_machine() in core qdev.c
blame me for adding one there. Which were hacks so we won't
have to re-factor core qdev code. But that doesn't justify adding more.)

This patch is an interim one and later in 25/28
device_post_init() content is moved to a more generic compat interface
implementation. That intended for use with types derived from Object
(i.e. not only qdev stuff). Hence I'd like to decouple it from
machine as a standalone feature as much as possible. So that
machine (or whatever else) will opt in in using facility.

> The previous code was clearer and easier to follow, and wasn't
> sensitive to subtle changes in initialization ordering (e.g. what
> happens if we create a device before *_register_compat_props() is
> called?).
Indeed It seems clearer to follow (that was my first impression as well),
until I went through whole series and thought it's basically the same,
So my choice was to use cleaner approach that we won't have to rewrite
in near future.

Thanks for bringing up ordering issue, we probably have one in this series.

But beside possible issue here, even with v3 variant we would still have
issues if objects are created before machine and accelerator instances are
created.
More correct way could be to register compat properties right away at
select_machine() time, we don't really need an instance for that, just access
to machine_class and do the same for 'accel' option. (that's probably doable
within this series) + some time later (on top of this series) a check that
no TYPE_COMPAT_PROPS were created at the moment compat properties are registered
so we would notice when we write something wrong.

If it's too much of refactoring (series is already big as it is), I would
compromise on qdev_get_machine() and adding TODO comments (or a series on top)
to make it correct and "race-resistant".

Marc are you sure it actually will work as expected with Object derived types?
   register_global_properties()
is being called after
   qemu_opts_foreach(... user_creatable_add_opts_foreach, object_create_initial ...)
so there is no compat properties registered when objects are created.
 
> >   
> > >  
> > > >
> > > >      qdev_prop_set_globals(DEVICE(obj));
> > > >  }  
> > > [...]
> > >
> > > --
> > > Eduardo  
>