This series delay timer_new from init into realize to avoid memleaks when we call 'device_list_properties'. And do timer_free only in s390x_cpu_finalize because it's hotplugable. However, mos6522_realize is never called at all due to the incorrect creation of it. So we fix the incorrect creation in mac_via/cuda/pmu first, then move the timer_new to mos6522_realize().
v1: - Delay timer_new() from init() to realize() to fix memleaks. v2: - Similarly to other cleanups, move timer_new into realize in target/s390x/cpu.c (Suggested by Philippe Mathieu-Daudé). - Send these two patches as a series instead of send each as a single patch but with wrong subject in v1. v3: - It's not valid in mos6522 if we move timer_new from init to realize, because it's never called at all. Thus, we remove null check in reset, and add calls to mos6522_realize() in mac_via_realize to make this move to be valid. - split patch by device to make it more clear. v4: - Also do timer_free on the error path in realize() and fix some coding style. Then use device_class_set_parent_unrealize to declare unrealize. - split the mos6522 patch into two, one to fix incorrect creation of mos6522, the other to fix memleak. v5: - Fix two other places where we create mos6522's subclasses but forgot to realize it(macio/cuda,macio/pmu). Otherwise, this will cause SEGVs during make check-qtest-ppc64. - Remove timer_del on the error path of s390x_cpu_realize() and simply use errp instead a temporary variable. Pan Nengyuan (4): s390x: fix memleaks in cpu_finalize mac_via: fix incorrect creation of mos6522 device in mac_via hw/misc/macio: fix incorrect creation of mos6522's subclasses hw/misc/mos6522: move timer_new from init() into realize() to avoid memleaks hw/misc/mac_via.c | 40 +++++++++++++++++++++++++++------------- hw/misc/macio/cuda.c | 11 +++++++++-- hw/misc/macio/pmu.c | 11 +++++++++-- hw/misc/mos6522.c | 6 ++++++ target/s390x/cpu-qom.h | 1 + target/s390x/cpu.c | 30 ++++++++++++++++++++++++++---- 6 files changed, 78 insertions(+), 21 deletions(-) -- 2.18.2