Hi Jason, I've prepared a POC of graceful switch to 'vhost off' if respective features are acked by the guest. Such a way we do not need to silently clear RSS and hash report features in case of 'vhost on'. Can you please review it and provide your feedback?
I think the only open question is what to do with cases of vhost-user and vhost-vdpa. https://github.com/qemu/qemu/pull/105 This pull request is for reviews only. Thanks in advance On Mon, Jan 18, 2021 at 5:16 AM Jason Wang <jasow...@redhat.com> wrote: > > > On 2021/1/17 下午5:04, Yuri Benditovich wrote: > > On Fri, Jan 15, 2021 at 9:20 AM Jason Wang<jasow...@redhat.com> wrote: > >> On 2021/1/15 上午5:16, Andrew Melnychenko wrote: > >>> From: Andrew<and...@daynix.com> > >>> > >>> When RSS is enabled the device tries to load the eBPF program > >>> to select RX virtqueue in the TUN. If eBPF can be loaded > >>> the RSS will function also with vhost (works with kernel 5.8 and later). > >>> Software RSS is used as a fallback with vhost=off when eBPF can't be > >>> loaded > >>> or when hash population requested by the guest. > >>> > >>> Signed-off-by: Yuri Benditovich<yuri.benditov...@daynix.com> > >>> Signed-off-by: Andrew Melnychenko<and...@daynix.com> > >>> --- > >>> hw/net/vhost_net.c | 2 + > >>> hw/net/virtio-net.c | 125 +++++++++++++++++++++++++++++++-- > >>> include/hw/virtio/virtio-net.h | 4 ++ > >>> net/vhost-vdpa.c | 2 + > >>> 4 files changed, 129 insertions(+), 4 deletions(-) > >>> > >>> diff --git a/hw/net/vhost_net.c b/hw/net/vhost_net.c > >>> index 24d555e764..16124f99c3 100644 > >>> --- a/hw/net/vhost_net.c > >>> +++ b/hw/net/vhost_net.c > >>> @@ -71,6 +71,8 @@ static const int user_feature_bits[] = { > >>> VIRTIO_NET_F_MTU, > >>> VIRTIO_F_IOMMU_PLATFORM, > >>> VIRTIO_F_RING_PACKED, > >>> + VIRTIO_NET_F_RSS, > >>> + VIRTIO_NET_F_HASH_REPORT, > >>> > >>> /* This bit implies RARP isn't sent by QEMU out of band */ > >>> VIRTIO_NET_F_GUEST_ANNOUNCE, > >>> diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c > >>> index 09ceb02c9d..37016fc73a 100644 > >>> --- a/hw/net/virtio-net.c > >>> +++ b/hw/net/virtio-net.c > >>> @@ -691,6 +691,19 @@ static void virtio_net_set_queues(VirtIONet *n) > >>> > >>> static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue); > >>> > >>> +static uint64_t fix_ebpf_vhost_features(uint64_t features) > >>> +{ > >>> + /* If vhost=on & CONFIG_EBPF doesn't set - disable RSS feature */ > >> I still think we should not clear feature silently. This may break > >> migraiton if the feature is cleared on destination. > > Do I understand it correctly that if we do not clear features silently > > and implement a graceful drop to vhost=off when we can't do what we > > need with vhost - then we do not need to add any migration blocker? > > > Yes. I think we won't go with migration blocker since we need support > migration in the end. > > Thanks > > > > >