On Wed, Mar 13, 2013 at 09:02:38AM +0100, Paolo Bonzini wrote: > Il 13/03/2013 04:13, Asias He ha scritto: > >> > This takes dev mutex on data path which will introduce > >> > contention esp for multiqueue. > > Yes, for now it is okay, but for concurrent execution of multiqueue it is > > really bad. > > > > By the way, what is the overhead of taking and releasing the > > vs->dev.mutex even if no one contents for it? Is this overhead gnorable. > > There is a possibility of cacheline ping-pong, but apart from that it's > ignorable. Ah, thanks! > >> > How about storing the endpoint as part of vq > >> > private data and protecting with vq mutex? > > > > Hmm, this makes sense, let's see how well it works. > > Then VHOST_SCSI_SET_ENDPOINT would have to go through all vqs, no? A > rwlock seems simpler. VHOST_SCSI_SET_ENDPOINT operation is not on the data path, it is fine to go through all vqs, it's just a loop. For the rwlock thing, let's discuss it on the other thread. > Paolo -- Asias -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
