I sent the exact same patches offline to some of the Cc'd, there's been no change but they have been tested by Jason now. I'll queue them up unless there are some comments.
What these patches do is arrive at a state where migrate to exec with vm stopped produces a stable result. With these patches applied it's stable, under the following conditions: 1. run with -rtc clock=vm otherwise the rtc clock keeps running with vm stopped. 2. kvm clock compiled out (-kvmclock in -cpu should have same efect but does not, this is a bug, and it breaks migration between hosts with and without kvmclock support in kernel). 3. ratelimit set to a very large value (like 100g) Tested-by: Jason Wang <jasow...@redhat.com> Michael S. Tsirkin (6): virtio-net: don't dma while vm is stopped cpus: flush all requests on each vm stop migration/savevm: no need to flush requests virtio-net: stop/start bh on vm start/stop migration: stable ram block ordering migration: allow rate > 4g arch_init.c | 35 +++++++++++++++++++++++++++++++++++ buffered_file.c | 9 ++++++--- cpu-common.h | 3 +++ cpus.c | 2 ++ exec.c | 24 ++++++++++++++++++++++-- hw/hw.h | 8 ++++---- hw/virtio-net.c | 39 +++++++++++++++++++++++++++++---------- kvm-all.c | 2 +- migration.c | 8 ++++---- savevm.c | 8 ++------ 10 files changed, 108 insertions(+), 30 deletions(-) -- 1.7.3.2.91.g446ac
