On 6/3/07 11:24, "John Levon" <levon@xxxxxxxxxxxxxxxxx> wrote:
> I'm unclear on why we have to have the mapping part of solving the
> xendev problem. Can't we just allocate a transaction as usual if we see
> an ID without a transaction existing, mark it as pre-fail, and when we
> see the transaction end, discard the copy as if it had raced?
>
> Surely the ID lookup is guaranteed to fail in the circumstances we care
> about (restore, xenstored restart)?
Xenstored allocates transaction ids, and it can reuse them after
restore/restart. So you'll be able to end up with multiple transaction
contexts in the guest who were all originally allocated the same transaction
id by xenstored. Of course all but one is certainly doomed, but you need to
disambiguate...
 -- Keir
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel