On Fri, Jun 16, 2017 at 07:04:27PM +0200, Maxime Coquelin wrote: > > > On 06/16/2017 05:19 PM, Michael S. Tsirkin wrote: > > On Fri, Jun 16, 2017 at 04:57:01PM +0800, Jason Wang wrote: > > > > > > > > > On 2017年06月16日 11:22, Michael S. Tsirkin wrote: > > > > > I think the issues can be solved by VIRTIO_F_MAX_CHAIN_SIZE. > > > > > > > > > > For now, how about splitting it into two series of patches: > > > > > 1) enable 1024 tx queue size for vhost-user, to let the users of > > > > > vhost-user > > > > > to easily use 1024 queue size. > > > > Fine with me. 1) will get property from user but override it on > > > > !vhost-user. Do we need a protocol flag? It seems prudent but we get > > > > back to cross-version migration issues that a04re still pending > > > > solution. > > > > Marc Andre, what's the status of that work? > > > > > > > > > 2) enable VIRTIO_F_MAX_CHAIN_SIZE, to enhance robustness. > > > > Rather, to support it for more backends. > > > > > > Ok, if we want to support different values of max chain size in the > > > future. > > > It would be problematic for migration of cross backends, consider the case > > > when migrating from 2048 (vhost-user) to 1024 (qemu/vhost-kernel). > > > > > > Thanks > > > > That's already a problem, and it's growing with each new feature. > > Maxime looked at supporting vhost-user backends cross-version migration, > > I think we must merge some solution sooner rather than later, preferably > > by the next release. > > > > Maxime, any update here? Do we need a meeting to reach consensus? > > No update, I haven't found time to progress on the topic yet. > > For those who aren't aware of my initial proposal, you may find it here: > https://www.spinics.net/linux/fedora/libvir/msg142668.html > > If my understanding is correct, you were concerned about the complexity of > my > proposal which involved too many layers. Your suggestion was to have a tool > provided with qemu that would connect to vhost-user socket and query the > backend capabilities. > I'm not 100% clear how it would work, as the trend is to start the backend > in > client mode, meaning QEMU creates the socket. In this case, should the tool > create the socket and management tool request the backend to connect to it? > > I think it could make sense to have a meeting, but maybe we should first > discuss the solutions on the list for efficiency. > > For the delivery, what is QEMU v2.10 planned release date? > > Note that my solution doesn't involve QEMU, so it would not be tight to QEMU > release date. But, that doesn't mean it would be delivered sooner than > your solution. > > Maxime
I'd say let's go with your proposal (Solution 3 above). -- MST
