On Mon, 08 Aug 2011 17:27:24 +0300 Avi Kivity <a...@redhat.com> wrote:
> On 08/08/2011 05:06 PM, Luiz Capitulino wrote: > > > > > > This is why I suggested a reference count. In this case, we can always > > > stop the guest "twice", because we don't lost information when we resume. > > > > I could give it a try in the near future, as I really think it's independent > > from this series, but I still don't understand what can stop an already > > stopped > > VM besides the user. This is a real question, is it really possible? > > > > If only the user can do that, then the refcount is overkill as just > > returning > > an error will do it. > > Most of the reasons in QemuState are asynchronous and can happen at any > time while the guest is running. Because they're asynchronous, they can > trigger before a monitor stop is issued but caught after it is processed. > > We could possibly synchronize during user stop, but that lets us capture > only the first non-user reason. Isn't it good enough? Another question: do you think that this problem is so relevant that we should include the fix in the first merge? > > And even if we return an error, that only pushes the refcounting to the > user; you probably don't want to return a "vm is already stopped" error > to the user, he'll just ask "why are you telling me that". >
