On 02/07/2012 10:18 AM, Laine Stump wrote:
In the meantime, we still should work on Dan's proposal to make locking within the driver object more fine grained.
If only I ever had enough 'round tuits'.
(NB: although I couldn't find any instance where qemuProcessKil() was
s/Kil()/Kill()/
called with no jobs active for the domain (or some other guarantee that the current thread had at least one refcount on the domain object), this patch still follows Eric's method of temporarily adding a ref prior to unlocking the domain object, because I couldn't convince myself 100% that this was the case.)
Yeah, and that's probably best for future-proofing against other changes in the code base. Another one of those projects where I wish I had more time is the introduction of a virObject 'base' class, with ref-counting done by lightweight atomic operations rather than heavy-weight mutex syscalls, which would be the sort of rewrite where we could easily be altering whether callers also hold a reference. -- Eric Blake eblake@redhat.com +1-919-301-3266 Libvirt virtualization library http://libvirt.org