Description of problem:

qemu-kvm segfaults when one tries to change a migration speed after a migration has finished (in any state). I consider this pretty harsh, since a small mistake can lead to a guest crash. To be honest, I have no idea how I never came across this before.


How reproducible:
100 %

Steps to Reproduce:
In the monitor of src machine:
1. migrate <params>
2. wait until it fails/finishes
3. migrate_set_speed <whatever>
  
Actual results:
crash

Expected results:
migrate_set_speed is ignored, since it makes no sense to set speed of a migration that does not exist.

Additional info:
The problem exists upstream too, and a patch was sent by me today.
The patch (if accepted in its current state) is pretty trivial, and do not introduce further risks. Given a harm/risk ratio, I say we should consider it for inclusion as soon as upstream takes it.