Package: qemu-system-x86
Version: 1:2.1+dfsg-12~bpo70+1
I have seen unresponsive VMs after live migration several times. CPU
load goes up to maximum, with no reaction at all on network or console.
There is a known bug that looks similar:
[PATCH] kvmclock: Ensure time in migration never goes backward
http://comments.gmane.org/gmane.comp.emulators.kvm.devel/121498
The corresponding patch was released with qemu 2.1.3:
http://lists.gnu.org/archive/html/qemu-stable/2015-01/msg00143.html
c35ba0d: kvmclock: Ensure time in migration never goes backward
(Alexander Graf)
It is this patch:
* kvmclock: Ensure time in migration never goes backward
http://git.qemu-project.org/?p=qemu.git;a=commit;h=9a48bcd1b82494671c111109b0eefdb882581499
I tried it and at first it looked very good. But obviously there is a
reason that it took so long for the patch to get released: some migrated
VMs died some hours later with:
qemu-system-x86_64: qemu-2.1+dfsg/hw/i386/kvm/clock.c:68:
kvmclock_current_nsec: Assertion `time.tsc_timestamp <= migration_tsc'
failed.
Looking further, some of the patches released with 2.1.3 are probably
related. I ended up using also these:
* Introduce cpu_clean_all_dirty
http://git.qemu-project.org/?p=qemu.git;a=commit;h=de9d61e83d43be9069e6646fa9d57a3f47779d28
* kvmclock: Ensure proper env->tsc value for kvmclock_current_nsec
http://git.qemu-project.org/?p=qemu.git;a=commit;h=317b0a6d8ba44e9bf8f9c3dbd776c4536843d82c
* kvmclock: Add comment explaining why we need cpu_clean_all_dirty()
http://git.qemu-project.org/?p=qemu.git;a=commit;h=1154d84dcc5f46e83db94281d071775819dd8884
I think this says the same:
https://bugzilla.redhat.com/show_bug.cgi?id=1109427
Again, I made some migration tests with these patches, and had no more
problems.
Cheers
Harry
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
