This patchset fixes up various bugs in loadvm/savevm for -M g3beige and
-M mac99 so that it is becomes possible to save and restore image snapshots.
The focus of this patchset is on -M g3beige since this matches the majority
of my test images, but there were some easy fixes to be made to -M mac99
at the same time.
With this patchset applied both -M g3beige and -M mac99 images can be
saved/restored whilst booted into OpenBIOS with no issues. I tested -M g3beige
with a paused, disk-inactive Darwin 6 image and was able to resume
successfully which was good enough for my needs.
I noticed some hangs can still occur when trying to restore an image
where the disk is active which makes me believe that there is still some
extra macio/dbdma state which needs to be included if someone is interested
enough to pursue this further.
Most of the patches are straightforward except for patch 4 which came out of
a discussion on-list between Alex and Paolo, and patch 5 which is a similar
error except this time for the MSR register. I suspect patch 5 can be
improved by someone with more PPC knowledge than myself.
Signed-off-by: Mark Cave-Ayland <address@hidden>
Mark Cave-Ayland (7):
macio.c: include parent PCIDevice state in VMStateDescription
adb.c: include ADBDevice parent state in KBDState and MouseState
cuda.c: include adb_poll_timer in VMStateDescription
ppc: move sdr1 value change detection logic to helper_store_sdr1()
ppc: force update of all msr bits in cpu_post_load
openpic: fix segfault on -M mac99 savevm
openpic: fix up loadvm under -M mac99
hw/input/adb.c | 21 +++++++++++++++++----
hw/intc/openpic.c | 10 ++++------
hw/misc/macio/cuda.c | 5 +++--
hw/misc/macio/macio.c | 24 ++++++++++++++++++++++++
target-ppc/machine.c | 8 +++++++-
target-ppc/misc_helper.c | 7 ++++++-
target-ppc/mmu_helper.c | 35 +++++++++++++++--------------------
7 files changed, 76 insertions(+), 34 deletions(-)
Ping? Particularly patches 4 and 5 will require some review as they fix general target-ppc bugs when loading/saving VM state and aren't just specific to the g3beige/mac99 machines.
I decided to try them out. Using Mac OS 10.2 in the guest, I opened up a word processor and saved the state of QEMU. I then quit QEMU and started it back up. When the state was loaded, the screen looked exactly how it should, but nothing was working. QEMU was frozen. It was using around 100% of the CPU time, so it might have been stuck in some kind of loop. Once when I tried loading a saved state, it did work. Haven't been able to reproduce it after several attempts.
I suggest you use a fully installed guest OS for testing. Debian Linux 5 is good.
Just to note, I used QEMU 2.2.0 for testing. |