I''ve just observed some very odd behaviour in our (essentially
identical
to Linux''s) xen_shutdown_handler():
xm save johnlev /var/tmp/img
xen_shutdown_handler: "suspend"
xen_shutdown_handler: ""
The first is the watch asking us to suspend. Then we get an EAGAIN and
loop around to try again, but this time the node is rewritten to be
"",
as we requested via xenbus_write(). As far as I can see it shouldn''t be
possible for our xenbus_write("") to have succeeded as part of the
failed transaction. Is it possible the EAGAIN is coming from the higher
levels somehow, rather than do_transaction_end() ?
Unfortunately I didn''t have tracing on and have only ever reproduced
this once, so can''t provide any more detail.
regards
john
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel
On 4/10/06 3:13 am, "John Levon" <levon@movementarian.org> wrote:> The first is the watch asking us to suspend. Then we get an EAGAIN and > loop around to try again, but this time the node is rewritten to be "", > as we requested via xenbus_write(). As far as I can see it shouldn''t be > possible for our xenbus_write("") to have succeeded as part of the > failed transaction. Is it possible the EAGAIN is coming from the higher > levels somehow, rather than do_transaction_end() ?No, the writes should be discarded of course. If xenstored ever returns EAGAIN it should mean that the transaction was discarded and not committed in part or whole. Could EAGAIN be coming from the xenbus layers in your guest OS? -- Keir _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Wed, Oct 04, 2006 at 08:48:52AM +0100, Keir Fraser wrote:> > The first is the watch asking us to suspend. Then we get an EAGAIN and > > loop around to try again, but this time the node is rewritten to be "", > > as we requested via xenbus_write(). As far as I can see it shouldn''t be > > possible for our xenbus_write("") to have succeeded as part of the > > failed transaction. Is it possible the EAGAIN is coming from the higher > > levels somehow, rather than do_transaction_end() ? > > No, the writes should be discarded of course.Indeed.> Could EAGAIN be coming from the xenbus layers in your guest OS?I certainly can''t see any path right now. I''m trying to reproduce again to confirm what''s happening... regards john _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel