On Mon, 25 Jan 2010 16:33:02 +0200 Avi Kivity <a...@redhat.com> wrote:
> On 01/25/2010 04:29 PM, Markus Armbruster wrote: > > > > I agree with Anthony that async message masking doesn't really affect > > the protocol proper. We could pretend it does so we can let protocol > > capability negotiation (which we need anyway) cover it. But I'm > > certainly fine with keeping it separate. > > > > Whether we call it protocol or not, the question whether we should > > permit changing the masks at any time is valid, I think. Permitting it > > adds a bit of conceptual complexity, as a command disabling reporting of > > an event can race with the event. But that's just giving clients some > > more rope. I'm fine with that. > > > > Without disagreeing with the rest (which means I'm just nit-picking), > there's no race. Once the command that disables an event report returns > to the caller, the event can no longer be reported. I wouldn't call it a race, but if you don't want an event you'll have to deal with it between mode change and masking. Not a big deal, only confirms that clients are required to know how to ignore events, even if masking is available (which I'm not going to introduce).
