Message ID | 20230202110058.130695-25-xuanzhuo@linux.alibaba.com (mailing list archive) |
---|---|
State | Changes Requested |
Delegated to: | Netdev Maintainers |
Headers | show |
Series | virtio-net: support AF_XDP zero copy | expand |
On Thu, Feb 02, 2023 at 07:00:49PM +0800, Xuan Zhuo wrote: > Since xsk's TX queue is consumed by TX NAPI, if sq is bound to xsk, then > we must stop tx napi from being disabled. > > Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com> > --- > drivers/net/virtio/main.c | 9 ++++++++- > 1 file changed, 8 insertions(+), 1 deletion(-) > > diff --git a/drivers/net/virtio/main.c b/drivers/net/virtio/main.c > index ed79e750bc6c..232cf151abff 100644 > --- a/drivers/net/virtio/main.c > +++ b/drivers/net/virtio/main.c > @@ -2728,8 +2728,15 @@ static int virtnet_set_coalesce(struct net_device *dev, > return ret; > > if (update_napi) { > - for (i = 0; i < vi->max_queue_pairs; i++) > + for (i = 0; i < vi->max_queue_pairs; i++) { > + /* xsk xmit depend on the tx napi. So if xsk is active, depends. > + * prevent modifications to tx napi. > + */ > + if (rtnl_dereference(vi->sq[i].xsk.pool)) > + continue; > + > vi->sq[i].napi.weight = napi_weight; I don't get it. changing napi weight does not work then. why is this ok? > + } > } > > return ret; > -- > 2.32.0.3.g01195cf9f
On Thu, 2 Feb 2023 12:25:59 -0500, "Michael S. Tsirkin" <mst@redhat.com> wrote: > On Thu, Feb 02, 2023 at 07:00:49PM +0800, Xuan Zhuo wrote: > > Since xsk's TX queue is consumed by TX NAPI, if sq is bound to xsk, then > > we must stop tx napi from being disabled. > > > > Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com> > > --- > > drivers/net/virtio/main.c | 9 ++++++++- > > 1 file changed, 8 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/net/virtio/main.c b/drivers/net/virtio/main.c > > index ed79e750bc6c..232cf151abff 100644 > > --- a/drivers/net/virtio/main.c > > +++ b/drivers/net/virtio/main.c > > @@ -2728,8 +2728,15 @@ static int virtnet_set_coalesce(struct net_device *dev, > > return ret; > > > > if (update_napi) { > > - for (i = 0; i < vi->max_queue_pairs; i++) > > + for (i = 0; i < vi->max_queue_pairs; i++) { > > + /* xsk xmit depend on the tx napi. So if xsk is active, > > depends. > > > + * prevent modifications to tx napi. > > + */ > > + if (rtnl_dereference(vi->sq[i].xsk.pool)) > > + continue; > > + > > vi->sq[i].napi.weight = napi_weight; > > I don't get it. > changing napi weight does not work then. > why is this ok? static void skb_xmit_done(struct virtqueue *vq) { struct virtnet_info *vi = vq->vdev->priv; struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; /* Suppress further interrupts. */ virtqueue_disable_cb(vq); if (napi->weight) virtqueue_napi_schedule(napi, vq); else /* We were probably waiting for more output buffers. */ netif_wake_subqueue(vi->dev, vq2txq(vq)); } If the weight is 0, tx napi will not be triggered again. Thanks. > > > > + } > > } > > > > return ret; > > -- > > 2.32.0.3.g01195cf9f >
On Fri, Feb 03, 2023 at 11:24:42AM +0800, Xuan Zhuo wrote: > On Thu, 2 Feb 2023 12:25:59 -0500, "Michael S. Tsirkin" <mst@redhat.com> wrote: > > On Thu, Feb 02, 2023 at 07:00:49PM +0800, Xuan Zhuo wrote: > > > Since xsk's TX queue is consumed by TX NAPI, if sq is bound to xsk, then > > > we must stop tx napi from being disabled. > > > > > > Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com> > > > --- > > > drivers/net/virtio/main.c | 9 ++++++++- > > > 1 file changed, 8 insertions(+), 1 deletion(-) > > > > > > diff --git a/drivers/net/virtio/main.c b/drivers/net/virtio/main.c > > > index ed79e750bc6c..232cf151abff 100644 > > > --- a/drivers/net/virtio/main.c > > > +++ b/drivers/net/virtio/main.c > > > @@ -2728,8 +2728,15 @@ static int virtnet_set_coalesce(struct net_device *dev, > > > return ret; > > > > > > if (update_napi) { > > > - for (i = 0; i < vi->max_queue_pairs; i++) > > > + for (i = 0; i < vi->max_queue_pairs; i++) { > > > + /* xsk xmit depend on the tx napi. So if xsk is active, > > > > depends. > > > > > + * prevent modifications to tx napi. > > > + */ > > > + if (rtnl_dereference(vi->sq[i].xsk.pool)) > > > + continue; > > > + > > > vi->sq[i].napi.weight = napi_weight; > > > > I don't get it. > > changing napi weight does not work then. > > why is this ok? > > > static void skb_xmit_done(struct virtqueue *vq) > { > struct virtnet_info *vi = vq->vdev->priv; > struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; > > /* Suppress further interrupts. */ > virtqueue_disable_cb(vq); > > if (napi->weight) > virtqueue_napi_schedule(napi, vq); > else > /* We were probably waiting for more output buffers. */ > netif_wake_subqueue(vi->dev, vq2txq(vq)); > } > > > If the weight is 0, tx napi will not be triggered again. > > Thanks. This needs more thought then. First ignoring what user is requesting is not nice. Second what if napi is first disabled and then xsk enabled? > > > > > > > + } > > > } > > > > > > return ret; > > > -- > > > 2.32.0.3.g01195cf9f > >
On Fri, 3 Feb 2023 03:33:41 -0500, "Michael S. Tsirkin" <mst@redhat.com> wrote: > On Fri, Feb 03, 2023 at 11:24:42AM +0800, Xuan Zhuo wrote: > > On Thu, 2 Feb 2023 12:25:59 -0500, "Michael S. Tsirkin" <mst@redhat.com> wrote: > > > On Thu, Feb 02, 2023 at 07:00:49PM +0800, Xuan Zhuo wrote: > > > > Since xsk's TX queue is consumed by TX NAPI, if sq is bound to xsk, then > > > > we must stop tx napi from being disabled. > > > > > > > > Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com> > > > > --- > > > > drivers/net/virtio/main.c | 9 ++++++++- > > > > 1 file changed, 8 insertions(+), 1 deletion(-) > > > > > > > > diff --git a/drivers/net/virtio/main.c b/drivers/net/virtio/main.c > > > > index ed79e750bc6c..232cf151abff 100644 > > > > --- a/drivers/net/virtio/main.c > > > > +++ b/drivers/net/virtio/main.c > > > > @@ -2728,8 +2728,15 @@ static int virtnet_set_coalesce(struct net_device *dev, > > > > return ret; > > > > > > > > if (update_napi) { > > > > - for (i = 0; i < vi->max_queue_pairs; i++) > > > > + for (i = 0; i < vi->max_queue_pairs; i++) { > > > > + /* xsk xmit depend on the tx napi. So if xsk is active, > > > > > > depends. > > > > > > > + * prevent modifications to tx napi. > > > > + */ > > > > + if (rtnl_dereference(vi->sq[i].xsk.pool)) > > > > + continue; > > > > + > > > > vi->sq[i].napi.weight = napi_weight; > > > > > > I don't get it. > > > changing napi weight does not work then. > > > why is this ok? > > > > > > static void skb_xmit_done(struct virtqueue *vq) > > { > > struct virtnet_info *vi = vq->vdev->priv; > > struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; > > > > /* Suppress further interrupts. */ > > virtqueue_disable_cb(vq); > > > > if (napi->weight) > > virtqueue_napi_schedule(napi, vq); > > else > > /* We were probably waiting for more output buffers. */ > > netif_wake_subqueue(vi->dev, vq2txq(vq)); > > } > > > > > > If the weight is 0, tx napi will not be triggered again. > > > > Thanks. > > This needs more thought then. First ignoring what user is requesting is > not nice. Maybe we should return an error. >Second what if napi is first disabled and then xsk enabled? static int virtnet_xsk_pool_enable(struct net_device *dev, struct xsk_buff_pool *pool, u16 qid) { struct virtnet_info *vi = netdev_priv(dev); struct receive_queue *rq; struct send_queue *sq; int err; if (qid >= vi->curr_queue_pairs) return -EINVAL; sq = &vi->sq[qid]; rq = &vi->rq[qid]; /* xsk zerocopy depend on the tx napi. * * All xsk packets are actually consumed and sent out from the xsk tx * queue under the tx napi mechanism. */ -> if (!sq->napi.weight) return -EPERM; Thanks. > > > > > > > > > > > > + } > > > > } > > > > > > > > return ret; > > > > -- > > > > 2.32.0.3.g01195cf9f > > > >
On Fri, Feb 03, 2023 at 04:49:16PM +0800, Xuan Zhuo wrote: > On Fri, 3 Feb 2023 03:33:41 -0500, "Michael S. Tsirkin" <mst@redhat.com> wrote: > > On Fri, Feb 03, 2023 at 11:24:42AM +0800, Xuan Zhuo wrote: > > > On Thu, 2 Feb 2023 12:25:59 -0500, "Michael S. Tsirkin" <mst@redhat.com> wrote: > > > > On Thu, Feb 02, 2023 at 07:00:49PM +0800, Xuan Zhuo wrote: > > > > > Since xsk's TX queue is consumed by TX NAPI, if sq is bound to xsk, then > > > > > we must stop tx napi from being disabled. > > > > > > > > > > Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com> > > > > > --- > > > > > drivers/net/virtio/main.c | 9 ++++++++- > > > > > 1 file changed, 8 insertions(+), 1 deletion(-) > > > > > > > > > > diff --git a/drivers/net/virtio/main.c b/drivers/net/virtio/main.c > > > > > index ed79e750bc6c..232cf151abff 100644 > > > > > --- a/drivers/net/virtio/main.c > > > > > +++ b/drivers/net/virtio/main.c > > > > > @@ -2728,8 +2728,15 @@ static int virtnet_set_coalesce(struct net_device *dev, > > > > > return ret; > > > > > > > > > > if (update_napi) { > > > > > - for (i = 0; i < vi->max_queue_pairs; i++) > > > > > + for (i = 0; i < vi->max_queue_pairs; i++) { > > > > > + /* xsk xmit depend on the tx napi. So if xsk is active, > > > > > > > > depends. > > > > > > > > > + * prevent modifications to tx napi. > > > > > + */ > > > > > + if (rtnl_dereference(vi->sq[i].xsk.pool)) > > > > > + continue; > > > > > + > > > > > vi->sq[i].napi.weight = napi_weight; > > > > > > > > I don't get it. > > > > changing napi weight does not work then. > > > > why is this ok? > > > > > > > > > static void skb_xmit_done(struct virtqueue *vq) > > > { > > > struct virtnet_info *vi = vq->vdev->priv; > > > struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; > > > > > > /* Suppress further interrupts. */ > > > virtqueue_disable_cb(vq); > > > > > > if (napi->weight) > > > virtqueue_napi_schedule(napi, vq); > > > else > > > /* We were probably waiting for more output buffers. */ > > > netif_wake_subqueue(vi->dev, vq2txq(vq)); > > > } > > > > > > > > > If the weight is 0, tx napi will not be triggered again. > > > > > > Thanks. > > > > This needs more thought then. First ignoring what user is requesting is > > not nice. > > Maybe we should return an error. maybe > >Second what if napi is first disabled and then xsk enabled? > > > static int virtnet_xsk_pool_enable(struct net_device *dev, > struct xsk_buff_pool *pool, > u16 qid) > { > struct virtnet_info *vi = netdev_priv(dev); > struct receive_queue *rq; > struct send_queue *sq; > int err; > > if (qid >= vi->curr_queue_pairs) > return -EINVAL; > > sq = &vi->sq[qid]; > rq = &vi->rq[qid]; > > /* xsk zerocopy depend on the tx napi. > * > * All xsk packets are actually consumed and sent out from the xsk tx > * queue under the tx napi mechanism. > */ > -> if (!sq->napi.weight) > return -EPERM; > > Thanks. > > > > > > > > > > > > > > > > > > > + } > > > > > } > > > > > > > > > > return ret; > > > > > -- > > > > > 2.32.0.3.g01195cf9f > > > > > >
diff --git a/drivers/net/virtio/main.c b/drivers/net/virtio/main.c index ed79e750bc6c..232cf151abff 100644 --- a/drivers/net/virtio/main.c +++ b/drivers/net/virtio/main.c @@ -2728,8 +2728,15 @@ static int virtnet_set_coalesce(struct net_device *dev, return ret; if (update_napi) { - for (i = 0; i < vi->max_queue_pairs; i++) + for (i = 0; i < vi->max_queue_pairs; i++) { + /* xsk xmit depend on the tx napi. So if xsk is active, + * prevent modifications to tx napi. + */ + if (rtnl_dereference(vi->sq[i].xsk.pool)) + continue; + vi->sq[i].napi.weight = napi_weight; + } } return ret;
Since xsk's TX queue is consumed by TX NAPI, if sq is bound to xsk, then we must stop tx napi from being disabled. Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com> --- drivers/net/virtio/main.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)