Re: [Qemu-devel] [PATCH v3 5/8] qdev: Register static properties as class properties

[Eduardo Habkost]

On Thu, Oct 27, 2016 at 05:30:30PM +0200, Igor Mammedov wrote:
[...]
> > > "make V=1 check" fails for me with this patch applied:
> > > 
> > > QTEST_QEMU_BINARY=aarch64-softmmu/qemu-system-aarch64 
> > > QTEST_QEMU_IMG=qemu-img MALLOC_PERTURB_=${MALLOC_PERTURB_:-$((RANDOM % 
> > > 255 + 1))} gtester -k --verbose -m=quick  tests/device-introspect-test 
> > > tests/qom-test
> > > TEST: tests/device-introspect-test... (pid=275141)
> > >   /aarch64/device/introspect/list:                                     OK
> > >   /aarch64/device/introspect/none:                                     OK
> > >   /aarch64/device/introspect/abstract:                                 OK
> > >   /aarch64/device/introspect/concrete:                                 
> > > Unexpected error in object_property_find() at 
> > > /home/imammedo/builds/qemu/qom/object.c:1002:
> > > Property '.disable_vnet_hdr' not found  
> > 
> > Oops! Caused by the e1000e PropertyInfo hack that I tried to
> > remove yesterday. See "e1000e: QOM property & configuration
> > cleanups".
> It's property registration ordering issue exposed by 3/8
> 
> simple this this series should include after 3/8:

I'm considering simply squashing the fix below in 3/8, what do
you think?

> 
> diff --git a/hw/net/e1000e.c b/hw/net/e1000e.c
> index 368d284..b26760d 100644
> --- a/hw/net/e1000e.c
> +++ b/hw/net/e1000e.c
> @@ -671,7 +671,6 @@ static void e1000e_class_init(ObjectClass *class, void 
> *data)
>      dc->desc = "Intel 82574L GbE Controller";
>      dc->reset = e1000e_qdev_reset;
>      dc->vmsd = &e1000e_vmstate;
> -    device_class_set_props(dc, e1000e_properties);
>  
>      e1000e_prop_disable_vnet = qdev_prop_uint8;
>      e1000e_prop_disable_vnet.description = "Do not use virtio headers, "
> @@ -683,6 +682,7 @@ static void e1000e_class_init(ObjectClass *class, void 
> *data)
>  
>      e1000e_prop_subsys = qdev_prop_uint16;
>      e1000e_prop_subsys.description = "PCI device Subsystem ID";
> +    device_class_set_props(dc, e1000e_properties);
>  
>      set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
>  }
> 
> or event better fix ordering issues first and then generate 3/8 on top of 
> that.
> I wonder how many similar ordering 3/8 causes.

I've grepped for PropertyInfo usage, and this seems to be the
only case where PropertyInfo structs are modified at runtime.

-- 
Eduardo