On Wed, 9 Apr 2014, Francois Gouget wrote: [...]
As kvm_adm and kvm_intel are kernel modules, the kernel version might actually be more important than the QEMU one. Indeed my latest tests seem to indicate that reverting the kernel from the 3.13 to 3.2.0 fixes the VM corruption issues (I also tested 3.14rc7 which did not help).
This is confirmed: * The 3.2.0 kernel does not corrupt the VMs but might be causing the VM host crash that happens once in a while * The 3.12, 3.13.1 and 3.14rc7 kernels corrupt the VMs in short order and thus are not usable.
Anyway, I have reverted the VM host to the 3.2.0 kernel and restored the build and Windows 2000 VMs. I'm also hoping the git apply failures will be fixed after doing tonight's Git update.
The tests also confirm that the Dr6 bug is present all the way to the 3.13.1 kernel and is fixed in 3.14rc7. Huw probably found the commit that fixed it so there's potentially hope of backporting it:
73aaf249ee2287b4686ff079dcbdbbb658156e64 http://o.cs.uvic.ca:20810/perl/cid.pl?cid=73aaf249ee2287b4686ff079dcbdbbb658...
I could theoretically restore all the VMs, redo their live snapshots for QEMU 1.7.0, probably also do a Windows update for XP and greater since the most up to date configurations are now essentially a year old, but that will take a day or two. The live snapshots would also have to be redone if we switch to Intel (but the poweredoff post-Windows update snapshots should be reusable).
The advantage of switching to Intel is that it seems to be more tested upstream, and that both Huw and I have Intel systems. So it let us detect and potentially fix such issues before changing the WineTestBot configuration. It may also fix the occasional host crashes. But this state of affairs is pretty disappointing.
So if the Intel config can be put together in the few days it may be best to restore a minimal set of VMs, and restore the full set after the new hardware has proven itself.