This patch fixes an assert in spice-server code on the
destination side of migration, that happens when
the MIGRATE_DATA message of the main channel arrives
before the (qemu-kvm's) virtio-serial-port device state.
For a more detailed information look at the patch
log message.
Some comments/concerns about this patch:
1. Some more testing needs to be done.
To test it, the MIGRATE_DATA message needs to arrive first.
One way to achieve that is to make the migration socket
between the two hosts slower.
Another is to add some sleeps in the other path of the code, which
is what I did.
2. I did encounter migration problems in my testing, with
upstream code.
Seems to me it is not due to this patch but to either
a different migration problem or the code changes made
to test the patch.
3. Theoretically, any virtual device attached to a
virtio-serial-bus port, which is a means of
communication between the client and the guest,
should behave the same way. But the implementation
of main channel (vdagent) is different than the
implementation of other spice channels that use
virtio-serial port.
Based on a patch by Christophe.
Uri Lublin (1):
migration: Don't assert() if MIGRATE_DATA comes before attaching the
agent
server/reds.c | 39 ++++++++++++++++++++++++++++-----------
1 file changed, 28 insertions(+), 11 deletions(-)
--
1.9.3
_______________________________________________
Spice-devel mailing list
Spice-devel@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/spice-devel