[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CACGkMEuQuzy9Mn1JN-phAW+bk64yZsM1xoKTKP+B9R8KP36K2w@mail.gmail.com>
Date: Wed, 8 May 2024 10:22:04 +0800
From: Jason Wang <jasowang@...hat.com>
To: Heng Qi <hengqi@...ux.alibaba.com>
Cc: netdev@...r.kernel.org, virtualization@...ts.linux.dev,
"Michael S . Tsirkin" <mst@...hat.com>, Xuan Zhuo <xuanzhuo@...ux.alibaba.com>,
Eric Dumazet <edumazet@...gle.com>, "David S . Miller" <davem@...emloft.net>,
Jakub Kicinski <kuba@...nel.org>, Paolo Abeni <pabeni@...hat.com>
Subject: Re: [PATCH net-next v2 2/2] virtio_net: get init coalesce value when probe
On Tue, May 7, 2024 at 3:01 PM Heng Qi <hengqi@...ux.alibaba.com> wrote:
>
> On Tue, 7 May 2024 14:24:12 +0800, Jason Wang <jasowang@...hat.com> wrote:
> > On Fri, Apr 26, 2024 at 2:54 PM Heng Qi <hengqi@...ux.alibaba.com> wrote:
> > >
> > > Currently, virtio-net lacks a way to obtain the default coalesce
> > > values of the device during the probe phase. That is, the device
> > > may have default experience values, but the user uses "ethtool -c"
> > > to query that the values are still 0.
> > >
> > > Therefore, we reuse VIRTIO_NET_CTRL_NOTF_COAL_VQ_GET to complete the goal.
> > >
> > > Signed-off-by: Heng Qi <hengqi@...ux.alibaba.com>
> > > ---
> > > drivers/net/virtio_net.c | 68 +++++++++++++++++++++++++++++++++++-----
> > > 1 file changed, 61 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > index 3bc9b1e621db..fe0c15819dd3 100644
> > > --- a/drivers/net/virtio_net.c
> > > +++ b/drivers/net/virtio_net.c
> > > @@ -4623,6 +4623,46 @@ static int virtnet_validate(struct virtio_device *vdev)
> > > return 0;
> > > }
> > >
> > > +static int virtnet_get_coal_init_value(struct virtnet_info *vi,
> > > + u16 _vqn, bool is_tx)
> > > +{
> > > + struct virtio_net_ctrl_coal *coal = &vi->ctrl->coal_vq.coal;
> > > + __le16 *vqn = &vi->ctrl->coal_vq.vqn;
> > > + struct scatterlist sgs_in, sgs_out;
> > > + u32 usecs, pkts, i;
> > > + bool ret;
> > > +
> > > + *vqn = cpu_to_le16(_vqn);
> > > +
> > > + sg_init_one(&sgs_out, vqn, sizeof(*vqn));
> > > + sg_init_one(&sgs_in, coal, sizeof(*coal));
> > > + ret = virtnet_send_command_reply(vi, VIRTIO_NET_CTRL_NOTF_COAL,
> > > + VIRTIO_NET_CTRL_NOTF_COAL_VQ_GET,
> > > + &sgs_out, &sgs_in);
> > > + if (!ret)
> > > + return ret;
> > > +
> > > + usecs = le32_to_cpu(coal->max_usecs);
> > > + pkts = le32_to_cpu(coal->max_packets);
> > > + if (is_tx) {
> > > + vi->intr_coal_tx.max_usecs = usecs;
> > > + vi->intr_coal_tx.max_packets = pkts;
> > > + for (i = 0; i < vi->max_queue_pairs; i++) {
> > > + vi->sq[i].intr_coal.max_usecs = usecs;
> > > + vi->sq[i].intr_coal.max_packets = pkts;
> > > + }
> > > + } else {
> > > + vi->intr_coal_rx.max_usecs = usecs;
> > > + vi->intr_coal_rx.max_packets = pkts;
> > > + for (i = 0; i < vi->max_queue_pairs; i++) {
> > > + vi->rq[i].intr_coal.max_usecs = usecs;
> > > + vi->rq[i].intr_coal.max_packets = pkts;
> > > + }
> > > + }
> > > +
> > > + return 0;
> > > +}
> > > +
> > > static bool virtnet_check_guest_gso(const struct virtnet_info *vi)
> > > {
> > > return virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) ||
> > > @@ -4885,13 +4925,6 @@ static int virtnet_probe(struct virtio_device *vdev)
> > > vi->intr_coal_tx.max_packets = 0;
> > > }
> > >
> > > - if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL)) {
> > > - /* The reason is the same as VIRTIO_NET_F_NOTF_COAL. */
> > > - for (i = 0; i < vi->max_queue_pairs; i++)
> > > - if (vi->sq[i].napi.weight)
> > > - vi->sq[i].intr_coal.max_packets = 1;
> > > - }
> > > -
> > > #ifdef CONFIG_SYSFS
> > > if (vi->mergeable_rx_bufs)
> > > dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
> > > @@ -4926,6 +4959,27 @@ static int virtnet_probe(struct virtio_device *vdev)
> > >
> > > virtio_device_ready(vdev);
> > >
> > > + if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL)) {
> > > + /* The reason is the same as VIRTIO_NET_F_NOTF_COAL. */
> > > + for (i = 0; i < vi->max_queue_pairs; i++)
> > > + if (vi->sq[i].napi.weight)
> > > + vi->sq[i].intr_coal.max_packets = 1;
> > > +
> > > + /* The loop exits if the default value from any
> > > + * queue is successfully read.
> > > + */
> >
> > So this assumes the default values are the same. Is this something
> > required by the spec? If not, we probably need to iterate all the
> > queues.
> >
>
> From internal practice, and from the default behavior of other existing drivers,
> the queues all have the same value at the beginning, so here it seems feasible
> that we get the value of queue 0 to represent the global value instead of using
> a loop.
Well, unless the spec says the values are equal, the driver needs to iterate.
>
> Moreover, obtaining the value once for each queue initially does not seem to be
> very friendly for devices with a large number of queues.
We probably don't care too much about the time spent on the probe. For
example, there would be a lot of registers read/write as well.
Thanks
>
> Thanks.
>
> > Thanks
> >
> >
> > > + for (i = 0; i < vi->max_queue_pairs; i++) {
> > > + err = virtnet_get_coal_init_value(vi, rxq2vq(i), false);
> > > + if (!err)
> > > + break;
> > > + }
> > > + for (i = 0; i < vi->max_queue_pairs; i++) {
> > > + err = virtnet_get_coal_init_value(vi, txq2vq(i), true);
> > > + if (!err)
> > > + break;
> > > + }
> > > + }
> > > +
> > > _virtnet_set_queues(vi, vi->curr_queue_pairs);
> > >
> > > /* a random MAC address has been assigned, notify the device.
> > > --
> > > 2.32.0.3.g01195cf9f
> > >
> >
>
Powered by blists - more mailing lists