Re: [Qemu-devel] [PATCH v3 17/19] target-i386: fix apic object leak when CPU is deleted

[Eduardo Habkost <ehabkost@redhat.com>]

On Wed, Jul 13, 2016 at 05:46:25PM +0200, Igor Mammedov wrote:
> On Wed, 13 Jul 2016 17:26:18 +0200
> Igor Mammedov <imammedo@redhat.com> wrote:
> 
> > On Wed, 13 Jul 2016 12:04:44 -0300
> > Eduardo Habkost <ehabkost@redhat.com> wrote:
> > 
> > > On Wed, Jul 06, 2016 at 08:20:53AM +0200, Igor Mammedov wrote:  
> > > > Signed-off-by: Igor Mammedov <imammedo@redhat.com>
> > > > ---
> > > >  target-i386/cpu.c | 1 +
> > > >  1 file changed, 1 insertion(+)
> > > > 
> > > > diff --git a/target-i386/cpu.c b/target-i386/cpu.c
> > > > index 04c0b79..2fa445d 100644
> > > > --- a/target-i386/cpu.c
> > > > +++ b/target-i386/cpu.c
> > > > @@ -2765,6 +2765,7 @@ static void x86_cpu_apic_create(X86CPU *cpu, Error **errp)
> > > >  
> > > >      object_property_add_child(OBJECT(cpu), "lapic",
> > > >                                OBJECT(cpu->apic_state), &error_abort);
> > > > +    object_unref(OBJECT(cpu->apic_state));    
> > > 
> > > What kind of event can trigger object_unparent() or
> > > object_del_property() on "lapic"? Can we guarantee that the child
> > > property will never be deleted by any other code, only by
> > > x86_cpu_unrealizefn() and object_finalize(cpu)?  
> > code path that triggers unparent of lapic implicitly is
> > cpu instance removal when it deletes all children.
> > 
> > So unless someone adds explicit lapic removal somewhere in target-i386/cpu.c
> Well, I've wrote nonsense here as I do remove child explicitly
> in x86_cpu_unrealizefn(), so it's fine to set cpu->apic_state to NULL
> as you suggest in 18/19.
> 
> The other way around might be call only apic_state->unrealize() explicitly
> from x86_cpu_unrealizefn() and let QOM do unparenting/finalizing
> automatically for us. I'd even prefer this one over the former.
> Which one would you prefer?

I'd prefer to delete the object on unrealize, because it is
created on realize.

> 
> > I don't see how it could be deleted by other code path.
> That point still stands.

I don't see it, either, but I was not 100% sure it was not
possible to trigger property removal from other code paths (e.g.
QMP commands). If you are sure it can't be triggered:

Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>

-- 
Eduardo