Re: [Xen-devel] [PATCH v5 1/2] libxl: Introduce functions to add and remove USB devices to an HVM guest

[George Dunlap <george.dunlap@xxxxxxxxxxxxx>]

On 19/04/13 11:50, Ian Campbell wrote:
> On Fri, 2013-04-19 at 11:18 +0100, George Dunlap wrote:
> > We could add an option that says "attempt to do the remove even if
> > you
> > don't see the device in xenstore" I suppose.  But I'm not really sure
> > how to document that in a useful way without basically describing the
> > internal mechanics of the command.
> >
> > There's nothing we can really do about usb-list though.
> You could write it to xenstore before attempting the qmp commands. That
> just relies on the qmp remove failing gracefully for devices which
> aren't actually attached.
qmp remove will just return an error if you attempt to remove a device 
id it doesn't know about.  I suppose that could work.  It's also a bit 
more robust against, say, the xenstore add failing for whatever reason.
> You might also be able to come up with some complicated, prepare, do it,
> commit in xenstore like approach -- but that's probably a) difficult (if
> not impossibly due to races) and b) unnecessary.
Well if the failure mode is the process executing the libxl function 
being killed at a random point, I don't think we can do a transaction 
thing -- it would have to coordinate between qmp and xenstore.  AFAIK 
qmp doesn't have any transaction support, so who is going to do the 
cleanup?  The process that was just killed?  If it could, it could just 
finish whatever it was doing anyway.
 -George

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel