Re: question on PV pass-through patch

[Yosuke Iwamatsu <y-iwamatsu@ab.jp.nec.com>]

Masroor Vettuparambil wrote:
>  
> I tried your patch with CONFIG_XEN_PCIDEV_BACKEND_VPCI.
> Detached the pci on the source and attached another slot to migrated
> domU on destination. 
> Migration is happening smoothly. thanks for your work.
> 
> a minor thing I noticed is, dmesg on domU shows the following warning
> during pci-detach,
> WARNING: g.e. still in use!
> WARNING: leaking g.e. and page still in use!

I'm not sure what this means, but it looks like something is wrong
with the disconnect procedure of pcifront/pciback.

> I wonder if we can achieve a 'live' migration of direct access, ie,
> without interrupting the network traffic?
> Something like creating a vif(network-attach) on domU for carrying the
> n/w traffic temporarily till the migration is over.
> After migration, move the n/w traffic back from vif to newly
> pci-attached NIC on the destination.
> Any thought?

Actually I've been thinking of "passthrough domain live migration"
using a multipath-driver which combines a passthrough device and
a virtual device. Below is an idea of live migrating a domU
which has passthrough network device attached.

- First create bond0 and add the passthrough interface to it.
- Usually we use the passthrough interface (fast path).
- Before migration, attach a vif to the domU and add the vif interface
   to bond0. Then remove the passthrough inteface from bond0 and do
   pci-detach. We can migrate without network disconnection because
   the vif is now enabled (slow path).
- After migration, do pci-attach, add passthrough interface to bond0,
   and disconnect the vif interface (switch back to fast path).

I hope this method works, but I haven't have an opportunity to try it
myself yet.

Thanks,
-------------------
Yosuke Iwamatsu
     NEC Corporation