| Message ID | 20240802-rsc-v1-1-2b607bd2f555@daynix.com |
|---|---|
| State | New |
| Headers | show |
| Series | vhost: Add VIRTIO_NET_F_RSC_EXT to vhost feature bits | expand |
At the moment kernel (vhost) and vhost user backends do not support this feature (and IMO they do not need to) In order to support it they need to implement a) coalescing of segmented TCP packets and b) population of respective fields in host-to-guest packets (number of coalesced segments and counter of duplicated acks) Just to remind - this feature is intended to be used without vhost and without migration Thanks Yuri On Fri, Aug 2, 2024 at 8:38 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > VIRTIO_NET_F_RSC_EXT is implemented in the rx data path, which vhost > implements, so vhost needs to support the feature if it is ever to be > enabled with vhost. The feature must be disabled otherwise. > > Fixes: 2974e916df87 ("virtio-net: support RSC v4/v6 tcp traffic for Windows HCK") > Reported-by: Jason Wang <jasowang@redhat.com> > Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > --- > hw/net/vhost_net.c | 2 ++ > net/vhost-vdpa.c | 1 + > 2 files changed, 3 insertions(+) > > diff --git a/hw/net/vhost_net.c b/hw/net/vhost_net.c > index a788e6937e03..dedf9ad7c242 100644 > --- a/hw/net/vhost_net.c > +++ b/hw/net/vhost_net.c > @@ -50,6 +50,7 @@ static const int kernel_feature_bits[] = { > VIRTIO_F_RING_RESET, > VIRTIO_F_IN_ORDER, > VIRTIO_F_NOTIFICATION_DATA, > + VIRTIO_NET_F_RSC_EXT, > VIRTIO_NET_F_HASH_REPORT, > VHOST_INVALID_FEATURE_BIT > }; > @@ -81,6 +82,7 @@ static const int user_feature_bits[] = { > VIRTIO_F_RING_RESET, > VIRTIO_F_IN_ORDER, > VIRTIO_NET_F_RSS, > + VIRTIO_NET_F_RSC_EXT, > VIRTIO_NET_F_HASH_REPORT, > VIRTIO_NET_F_GUEST_USO4, > VIRTIO_NET_F_GUEST_USO6, > diff --git a/net/vhost-vdpa.c b/net/vhost-vdpa.c > index 03457ead663a..46b02c50be8c 100644 > --- a/net/vhost-vdpa.c > +++ b/net/vhost-vdpa.c > @@ -88,6 +88,7 @@ const int vdpa_feature_bits[] = { > VIRTIO_NET_F_MQ, > VIRTIO_NET_F_MRG_RXBUF, > VIRTIO_NET_F_MTU, > + VIRTIO_NET_F_RSC_EXT, > VIRTIO_NET_F_RSS, > VIRTIO_NET_F_STATUS, > VIRTIO_RING_F_EVENT_IDX, > > --- > base-commit: 31669121a01a14732f57c49400bc239cf9fd505f > change-id: 20240802-rsc-e90fb452bd7f > > Best regards, > -- > Akihiko Odaki <akihiko.odaki@daynix.com> >
On Fri, Aug 02, 2024 at 10:34:37AM +0300, Yuri Benditovich wrote: > At the moment kernel (vhost) and vhost user backends do not support > this feature (and IMO they do not need to) > In order to support it they need to implement a) coalescing of > segmented TCP packets and b) population of respective fields in > host-to-guest packets (number of coalesced segments and counter of > duplicated acks) > Just to remind - this feature is intended to be used without vhost and > without migration > > Thanks > Yuri it is off ny default so no big deal ... if it's incompatiblr with vhost/migration I guess it is a good idea to make sure users do not enable it like this.
On Fri, Aug 2, 2024 at 1:38 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > VIRTIO_NET_F_RSC_EXT is implemented in the rx data path, which vhost > implements, so vhost needs to support the feature if it is ever to be > enabled with vhost. The feature must be disabled otherwise. > > Fixes: 2974e916df87 ("virtio-net: support RSC v4/v6 tcp traffic for Windows HCK") > Reported-by: Jason Wang <jasowang@redhat.com> > Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > --- Acked-by: Jason Wang <jasowang@redhat.com> Thanks
diff --git a/hw/net/vhost_net.c b/hw/net/vhost_net.c index a788e6937e03..dedf9ad7c242 100644 --- a/hw/net/vhost_net.c +++ b/hw/net/vhost_net.c @@ -50,6 +50,7 @@ static const int kernel_feature_bits[] = { VIRTIO_F_RING_RESET, VIRTIO_F_IN_ORDER, VIRTIO_F_NOTIFICATION_DATA, + VIRTIO_NET_F_RSC_EXT, VIRTIO_NET_F_HASH_REPORT, VHOST_INVALID_FEATURE_BIT }; @@ -81,6 +82,7 @@ static const int user_feature_bits[] = { VIRTIO_F_RING_RESET, VIRTIO_F_IN_ORDER, VIRTIO_NET_F_RSS, + VIRTIO_NET_F_RSC_EXT, VIRTIO_NET_F_HASH_REPORT, VIRTIO_NET_F_GUEST_USO4, VIRTIO_NET_F_GUEST_USO6, diff --git a/net/vhost-vdpa.c b/net/vhost-vdpa.c index 03457ead663a..46b02c50be8c 100644 --- a/net/vhost-vdpa.c +++ b/net/vhost-vdpa.c @@ -88,6 +88,7 @@ const int vdpa_feature_bits[] = { VIRTIO_NET_F_MQ, VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_MTU, + VIRTIO_NET_F_RSC_EXT, VIRTIO_NET_F_RSS, VIRTIO_NET_F_STATUS, VIRTIO_RING_F_EVENT_IDX,
VIRTIO_NET_F_RSC_EXT is implemented in the rx data path, which vhost implements, so vhost needs to support the feature if it is ever to be enabled with vhost. The feature must be disabled otherwise. Fixes: 2974e916df87 ("virtio-net: support RSC v4/v6 tcp traffic for Windows HCK") Reported-by: Jason Wang <jasowang@redhat.com> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> --- hw/net/vhost_net.c | 2 ++ net/vhost-vdpa.c | 1 + 2 files changed, 3 insertions(+) --- base-commit: 31669121a01a14732f57c49400bc239cf9fd505f change-id: 20240802-rsc-e90fb452bd7f Best regards,