On Mon, Apr 23, 2012 at 2:19 PM, Michael S. Tsirkin <m...@redhat.com> wrote: > This is a follow-up to my previous patch: it turns > out that a single mb() isn't sufficient as network > loss could still be triggered under stress. > > Patch 1 is repost of v1. > > The following two patches fix more races found > by code inspection and comparison with vhost > in kernel. After applying these > patches, no more network loss was observed. > > Michael S. Tsirkin (3): > virtio: add missing mb() on notification > virtio: add missing mb() on enable notification > virtio: order index/descriptor reads > > hw/virtio.c | 11 +++++++++++ > qemu-barrier.h | 28 +++++++++++++++++++++++++--- > 2 files changed, 36 insertions(+), 3 deletions(-)
Reviewed-by: Stefan Hajnoczi <stefa...@linux.vnet.ibm.com>