Re: Memory allocation with rebooted HVM

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Tue, Oct 13, 2009 at 07:30:32PM +0100, Stefano Stabellini wrote:
> On Tue, 13 Oct 2009, Keir Fraser wrote:
> > On 13/10/2009 19:13, "Stefano Stabellini" <stefano.stabellini@eu.citrix.com>
> > wrote:
> > 
> > >> This seems to have been triggered just by that one domU being restarted.
> > >> I've seen this before as well, when a 4G domU tried to restart and hosed a
> > >> box that only had 3G of memory free (on top of the 4G it was using).
> > > 
> > > It seems to me that the new domain is always created after the old one
> > > has been destroyed with the exception that the old stubdom may still be
> > > alive but one stubdom uses only 32MB of ram, therefore cannot be the one
> > > preventing you from restarting the domain, especially in the case above
> > > where you had 3G free.
> > 
> > How much guest memory might the stubdom map? Could the stubdom be preventing
> > the guest's memory from being freed, by holding references to the pages?
> > 
> 
> Yes it could, this can also be the reason why he was able to reproduce
> this problem only few times (it depends on how much memory the stubdom has
> mapped, and that can vary from case to case).

We had encountered this problem in the past with blkback and with the SCSI disks
being iSCSI. The page's had a page-reference that would never decrement and 
the guest would stay in its zombie state. I've posted a patch some-time ago:

http://lists.xensource.com/archives/html/xen-devel/2009-09/msg00561.html

that can help troubleshoot if this is indeed this type of failure.