From: Avi Kivity Subject: Re: [PATCH 0/16 v6] PCI: Linux kernel SR-IOV support Date: Sun, 09 Nov 2008 14:53:05 +0200 Message-ID: <4916DD31.4000503@redhat.com> References: <20081106154351.GA30459@kroah.com> <894107.30288.qm@web45108.mail.sp1.yahoo.com> <20081106164919.GA4099@kroah.com> <0199E0D51A61344794750DC57738F58E5E26F996C4@GVW1118EXC.americas.hpqcorp.net> <20081106183630.GD11773@parisc-linux.org> <491371F0.7020805@codemonkey.ws> <87d4h7pnnm.fsf@basil.nowhere.org> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: Anthony Liguori , Matthew Wilcox , "Fischer, Anna" , Greg KH , H L , "randy.dunlap@oracle.com" , "grundler@parisc-linux.org" , "Chiang, Alexander" , "linux-pci@vger.kernel.org" , "rdreier@cisco.com" , "linux-kernel@vger.kernel.org" , "jbarnes@virtuousgeek.org" , "virtualization@lists.linux-foundation.org" , "kvm@vger.kernel.org" , "mingo@elte.hu" To: Andi Kleen Return-path: In-Reply-To: <87d4h7pnnm.fsf@basil.nowhere.org> Sender: linux-pci-owner@vger.kernel.org List-Id: kvm.vger.kernel.org Andi Kleen wrote: > Anthony Liguori writes: > >> What we would rather do in KVM, is have the VFs appear in the host as >> standard network devices. We would then like to back our existing PV >> driver to this VF directly bypassing the host networking stack. A key >> feature here is being able to fill the VF's receive queue with guest >> memory instead of host kernel memory so that you can get zero-copy >> receive traffic. This will perform just as well as doing passthrough >> (at least) and avoid all that ugliness of dealing with SR-IOV in the >> guest. >> > > But you shift a lot of ugliness into the host network stack again. > Not sure that is a good trade off. > The net effect will be positive. We will finally have aio networking from userspace (can send process memory without resorting to sendfile()), and we'll be able to assign a queue to a process (which will enable all sorts of interesting high performance things; basically VJ channels without kernel involvement). > Also it would always require context switches and I believe one > of the reasons for the PV/VF model is very low latency IO and having > heavyweight switches to the host and back would be against that. > It's true that latency would suffer (or alternatively cpu consumption would increase). -- error compiling committee.c: too many arguments to function