Re: [RFC PATCH 0/1] Removing RAMBlocks during migration

[Juan Quintela]

"Michael S. Tsirkin" <address@hidden> wrote:
> On Mon, Dec 09, 2019 at 10:41:01AM +0300, Yury Kotov wrote:
>> Hi,
>> 
>> I found that it's possible to remove a RAMBlock during migration.
>> E.g. device hot-unplugging initiated by a guest (how to reproduce is below).
>> And I want to clarify whether RAMBlock removing (or even adding) during
>> migration is valid operation or it's a bug.
>
> There's a very basic problem though: list of RAMBlock's on source and
> destination must match otherwise destination will be confused.
>
> It is probably fixable: keep a fake RAMBlock around until migration is
> complete, and send some kind of "RAMBlock removed" message to
> destination so it knows to remove it there as well.

Do we have the data to know that a device is unplugged?
I think that it would be just easier just queue that command.  Let the
guest unplug whatever it wants, and on destination, once that we know
that we have finish migration, we just do the rest of the staff.

As stated on the thread, we really want _not_ to have hotplug/unplug
during migration, but if a guest can initiate one, we can only "delay"
the bits that happen inside qemu.

Later, Juan.