From mboxrd@z Thu Jan 1 00:00:00 1970 From: Mark McLoughlin Subject: [PATCH 0/0][RFC] KVM use of vringfd Date: Fri, 13 Jun 2008 14:57:56 +0100 Message-ID: <1213365481-23460-1-git-send-email-markmc@redhat.com> Cc: kvm@vger.kernel.org To: Anthony Liguori , Avi Kivity , Rusty Russell Return-path: Received: from mail01.svc.cra.dublin.eircom.net ([159.134.118.17]:40037 "HELO mail01.svc.cra.dublin.eircom.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with SMTP id S1756026AbYFMN6E (ORCPT ); Fri, 13 Jun 2008 09:58:04 -0400 Sender: kvm-owner@vger.kernel.org List-ID: Hey, Since this came up at the KVM forum, I thought I'd post these patches here for discussion. The first issue with making KVM use vringfd is that the mmap() interface doesn't work for KVM since it is the guest that allocates the ring descriptor, so it doesn't make much sense for the host to remap those pages. The second issue is a little more tricky; QEMU's networking implementation is based around the notion of connecting a number of interfaces to a VLAN. So, in less common cases where you don't just have a tap interface and a virtio interface connected to the VLAN, you can't take just lguest's simple approach of sharing the same vrings between the tap driver in the host and the virtio driver in the guest and having the host userspace notify each side of buffers as they come and go. I know Anthony has ideas here, so I'll just let him outline them. Cheers, Mark.