Hello Anthony, Here's a rebased version of device-only QOM realization, otherwise unchanged. This is based on master now and drops all general-purpose qdev cleanups that potentially conflict with Paolo's reference counting change proposal. It will trivially conflict with my qdev unparenting fix, pending to be added to the qom-cpu queue (touches k->unparent in class_init).
Same as v1, this is not the big throw of a recursive, central realization model. The immediate purpose of this series is to allow CPUs, which partially already have realizefns prepared, to adjust to the new DeviceState signature (being a device on qom-cpu branch) and hook them up to DeviceClass::realize rather than exposing as a function through cpu.h. Please review and apply to qemu.git. As discussed in the last cover letter, doing realization at DeviceState level does not allow for trivial [un]realize_children as proposed by Paolo for Object. An idea how to implement this as a follow-up would be to do a deep search for child<> properties dynamic_cast'able to DeviceState. Or to simply try setting a "realized" property and it that fails check for children (but how to differ between "realized" that failed and "realized" that doesn't exist then?). Available from: https://github.com/afaerber/qemu-cpu/commits/realize-qdev.v2 git://github.com/afaerber/qemu-cpu.git realize-qdev.v2 Regards, Andreas Cc: Anthony Liguori <anth...@codemonkey.ws> Cc: Paolo Bonzini <pbonz...@redhat.com> Cc: Eduardo Habkost <ehabk...@redhat.com> Cc: Igor Mammedov <imamm...@redhat.com> Cc: Peter Maydell <peter.mayd...@linaro.org> Andreas Färber (2): qdev: Fold state enum into bool realized qdev: Prepare "realized" property hw/qdev-addr.c | 2 +- hw/qdev-core.h | 11 ++--- hw/qdev-properties-system.c | 4 +- hw/qdev-properties.c | 24 +++++----- hw/qdev.c | 106 +++++++++++++++++++++++++++++++------------ 5 Dateien geändert, 97 Zeilen hinzugefügt(+), 50 Zeilen entfernt(-) -- 1.7.10.4
