Am 21.01.2013 13:30, schrieb Paolo Bonzini: > BusState subclasses need to do their own allocation because > qbus_create_inplace calls object_initialize (which wipes out the > "free" callback). This patch separates the initialization of the object > (object_initialize) from its insertion in the qdev tree (qbus_realize); to > do so, it moves the remaining bits of qbus_create_inplace to qbus_realize > and export it as qbus_init. > > Signed-off-by: Paolo Bonzini <pbonz...@redhat.com>
I think I left some comments on v1: Would it be possible to leave a bus_initialize() function (without q ;)) in place that calls object_initialize() plus the-artist-formerly-known-as-qbus_realize(), shared between object_new() and object_initialize()? ->free was always set afterwards. The issue I am trying to contain here is a surge of *_init functions beyond class_init, instance_init, DeviceClass::init. Sticking to the QOM naming of having *bus_initialize() and *bus_new() would address that. Maybe if we reorder the two patches, dropping the use of g_malloc0() first? Currently care needs to be taken with the in-place bus initialization functions to not apply PCI_BUS() etc. on the uninitialized variable. Having a pci_bus_initialize(void *, ...) -> bus_initialize(void *, ...) -> object_initialize(void *, ...) call chain would solve that. Andreas -- SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg
