On 09/14/2012 03:30 PM, Stefan Hajnoczi wrote:
On Fri, Sep 14, 2012 at 2:34 AM, Amos Kong<ak...@redhat.com> wrote:
On 13/09/12 20:29, Stefan Hajnoczi wrote:
On Thu, Sep 13, 2012 at 9:51 AM, Amos Kong<ak...@redhat.com> wrote:
Regarding migration: do we migrate the NetClient->link_down field? If
we only migrate the status register value then the link may actually
be up at the net.c level.
I tried to add 'MediaStatus' to 'struct RTL8139State', and update
'VMStateDescription vmstate_rtl8139', then the value of MediaStatus
will be migrated.
But the idea in v2 is better.
Migrating the NIC's media status is not enough. Above I asked about
migrating nc->link_down, which determines whether net.c delivers
packets or drops them.
Your patch migrates the NIC's media status but I believe nc->link_down
isn't being migrated and the guest will therefore receive packets from
the host! This could lead to unexpected results since the guest
thinks the link is down.
It's not a bug in your patch, but a larger issue that needs to be
addressed for all NICs that support migration. (Unless I missed the
code that will migrate link_down.)
Stefan
A possible solution is to infer the nc->link_down from the media status
register in the destination. It works without adding more codes net.c
but need model specific callback functions.
