On 15.06.2018 11:16, Igor Mammedov wrote: > On Thu, 14 Jun 2018 16:10:14 +0200 > David Hildenbrand <da...@redhat.com> wrote: > >> On 14.06.2018 15:24, Igor Mammedov wrote: >>> On Wed, 13 Jun 2018 16:03:35 +0200 >>> David Hildenbrand <da...@redhat.com> wrote: >>> >>>> On 13.06.2018 14:56, Igor Mammedov wrote: >>>>> On Mon, 11 Jun 2018 14:16:50 +0200 >>>>> David Hildenbrand <da...@redhat.com> wrote: >>>>> >>>>>> "size" should not be queried before the device was realized. Let' make >>>>>> that explicit. >>>>>> >>>>>> Signed-off-by: David Hildenbrand <da...@redhat.com> >>>>> It's generic property getter, it should work even before realize is >>>>> called. >>>> >>>> That's the point, as we can see in NVDIMM code , it is *not* a generic >>>> property getter. >>> it is added by >>> object_property_add(obj, PC_DIMM_SIZE_PROP, "uint64", pc_dimm_get_size, >>> >>> NULL, NULL, NULL, &error_abort); >>> hence it's generic QOM property as opposed to memory device class specific >>> callbacks. >>> >>>> >>>>> >>>>> if issues described in 5/11 are properly fixed, this patch won't be >>>>> needed. >>>>> >>>>> So drop this patch >>>>> >>>>> >>>>>> --- >>>>>> hw/mem/pc-dimm.c | 6 ++++++ >>>>>> 1 file changed, 6 insertions(+) >>>>>> >>>>>> diff --git a/hw/mem/pc-dimm.c b/hw/mem/pc-dimm.c >>>>>> index 86fbcf2d0c..5294734529 100644 >>>>>> --- a/hw/mem/pc-dimm.c >>>>>> +++ b/hw/mem/pc-dimm.c >>>>>> @@ -166,6 +166,12 @@ static void pc_dimm_get_size(Object *obj, Visitor >>>>>> *v, const char *name, >>>>>> PCDIMMDevice *dimm = PC_DIMM(obj); >>>>>> PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(obj); >>>>>> >>>>>> + if (!DEVICE(obj)->realized) { >>>>>> + error_setg(errp, "Property \"%s\" not accessible before >>>>>> realized", >>>>>> + name); >>>>>> + return; >>>>>> + } >>>>>> + >>>>>> mr = ddc->get_memory_region(dimm, errp); >>> just make ddc->get_memory_region return error if 'memdev' isn't set yet >>> and add local_error handling here as proper property assessor should do. >>> >> >> Does not help for NVDIMM. > NVDIMM should be fixed as it's broken now and works just by accident. > I don't like tailoring infrastructure changes to a particular broken > device especially when the later could be made to work within current > framework. >
Yes, but I will go the path of static properties for NVDIMM instead. Trying to initialize the memory region in a property setter ("label-size") and then bailing out because another property (memdev) is not set looks strange. I will do checks+initialization in realize() and in get_memory_region(). This looks better in my point of view and looks like a nice NVDIMM refactoring (at least judging from the patches I have on my list now :) ). -- Thanks, David / dhildenb