[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH] virtio-net: set the max of queue size to 4096
From: |
Michael S. Tsirkin |
Subject: |
Re: [PATCH] virtio-net: set the max of queue size to 4096 |
Date: |
Wed, 21 Sep 2022 02:24:16 -0400 |
On Wed, Sep 21, 2022 at 02:15:41PM +0800, Jason Wang wrote:
> On Wed, Sep 21, 2022 at 11:57 AM Michael S. Tsirkin <mst@redhat.com> wrote:
> >
> > On Wed, Sep 21, 2022 at 10:59:50AM +0800, Jason Wang wrote:
> > > On Tue, Sep 20, 2022 at 8:59 PM Michael S. Tsirkin <mst@redhat.com> wrote:
> > > >
> > > > On Tue, Sep 20, 2022 at 10:03:23AM +0800, Jason Wang wrote:
> > > > > On Tue, Sep 20, 2022 at 9:38 AM Jason Wang <jasowang@redhat.com>
> > > > > wrote:
> > > > > >
> > > > > > On Tue, Sep 20, 2022 at 9:10 AM liuhaiwei <liuhaiwei9699@126.com>
> > > > > > wrote:
> > > > > > >
> > > > > > > From: liuhaiwei <liuhaiwei@inspur.com>
> > > > > > >
> > > > > > > the limit of maximum of rx_queue_size and tx_queue to 1024 is so
> > > > > > > small as to affect our network performance when using the
> > > > > > > virtio-net and vhost ,
> > > > > > > we cannot set the maximum size beyond 1k.
> > > > > > > why not enlarge the maximum size (such as 4096) when using the
> > > > > > > vhost backend?
> > > > > >
> > > > > > As Michael mentioned, there's a limitation in the kernel UIO_MAXIOV.
> > > > > > We need to find way to overcome that limit first.
> > > > >
> > > > > Btw, this probably means the skb needs to be built by vhost-net
> > > > > itself, instead of tuntap.
> > > > >
> > > > > Thanks
> > > >
> > > > That might help vhost-net but it won't help virtio-net.
> > > >
> > > > IMO the right fix is to add a separate limit on s/g size
> > > > to the spec. Block and scsi already have it, seems
> > > > reasonable to add it for others too.
> > >
> > > I wonder if it would be simpler to tie the limit to the virtqueue size.
> >
> > Simpler but wrong I think. As another example for RX we do not need s/g
> > at all with mergeable.
>
> But this is not forbidden by the spec if I was not wrong.
Yes, it is forbidden for device not to support S/G on any queues.
There's no way to express such a device limitation in the spec.
> > At the same time deep queues are needed
> > to avoid underruns.
> >
> >
> >
> > > Having individual limits seems to complicate the migration compatibility
> > > anyhow.
> > >
> > > Thanks
> >
> > I don't see how it's materially different from queue size.
>
> Yes.
>
> > We need interfaces for that, that we lack now, userspace is
> > expected to guess a subset available on both ends.
> >
>
> Ok.
>
> Thanks
>
> > > >
> > > >
> > > >
> > > > > >
> > > > > > Thanks
> > > > > >
> > > > > > >
> > > > > > > Signed-off-by: liuhaiwei <liuhaiwei9699@126.com>
> > > > > > > Signed-off-by: liuhaiwei <liuhaiwei@inspur.com>
> > > > > > > ---
> > > > > > > hw/net/virtio-net.c | 47
> > > > > > > +++++++++++++++++++++++++++-----------
> > > > > > > hw/virtio/virtio.c | 8 +++++--
> > > > > > > include/hw/virtio/virtio.h | 1 +
> > > > > > > 3 files changed, 41 insertions(+), 15 deletions(-)
> > > > > > >
> > > > > > > diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
> > > > > > > index dd0d056fde..4b56484855 100644
> > > > > > > --- a/hw/net/virtio-net.c
> > > > > > > +++ b/hw/net/virtio-net.c
> > > > > > > @@ -52,12 +52,11 @@
> > > > > > > #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */
> > > > > > >
> > > > > > > /* previously fixed value */
> > > > > > > -#define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
> > > > > > > -#define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
> > > > > > > +#define VIRTIO_NET_VHOST_USER_DEFAULT_SIZE 2048
> > > > > > >
> > > > > > > /* for now, only allow larger queue_pairs; with virtio-1, guest
> > > > > > > can downsize */
> > > > > > > -#define VIRTIO_NET_RX_QUEUE_MIN_SIZE
> > > > > > > VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE
> > > > > > > -#define VIRTIO_NET_TX_QUEUE_MIN_SIZE
> > > > > > > VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE
> > > > > > > +#define VIRTIO_NET_RX_QUEUE_MIN_SIZE 256
> > > > > > > +#define VIRTIO_NET_TX_QUEUE_MIN_SIZE 256
> > > > > > >
> > > > > > > #define VIRTIO_NET_IP4_ADDR_SIZE 8 /* ipv4 saddr +
> > > > > > > daddr */
> > > > > > >
> > > > > > > @@ -594,6 +593,28 @@ static int peer_has_ufo(VirtIONet *n)
> > > > > > > return n->has_ufo;
> > > > > > > }
> > > > > > >
> > > > > > > +static void virtio_net_set_default_queue_size(VirtIONet *n)
> > > > > > > +{
> > > > > > > + NetClientState *peer = n->nic_conf.peers.ncs[0];
> > > > > > > +
> > > > > > > + /* Default value is 0 if not set */
> > > > > > > + if (n->net_conf.rx_queue_size == 0) {
> > > > > > > + if (peer && peer->info->type ==
> > > > > > > NET_CLIENT_DRIVER_VHOST_USER) {
> > > > > > > + n->net_conf.rx_queue_size =
> > > > > > > VIRTIO_NET_VHOST_USER_DEFAULT_SIZE;
> > > > > > > + } else {
> > > > > > > + n->net_conf.rx_queue_size = VIRTIO_NET_VQ_MAX_SIZE;
> > > > > > > + }
> > > > > > > + }
> > > > > > > +
> > > > > > > + if (n->net_conf.tx_queue_size == 0) {
> > > > > > > + if (peer && peer->info->type ==
> > > > > > > NET_CLIENT_DRIVER_VHOST_USER) {
> > > > > > > + n->net_conf.tx_queue_size =
> > > > > > > VIRTIO_NET_VHOST_USER_DEFAULT_SIZE;
> > > > > > > + } else {
> > > > > > > + n->net_conf.tx_queue_size = VIRTIO_NET_VQ_MAX_SIZE;
> > > > > > > + }
> > > > > > > + }
> > > > > > > +}
> > > > > > > +
> > > > > > > static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int
> > > > > > > mergeable_rx_bufs,
> > > > > > > int version_1, int
> > > > > > > hash_report)
> > > > > > > {
> > > > > > > @@ -633,7 +654,7 @@ static int
> > > > > > > virtio_net_max_tx_queue_size(VirtIONet *n)
> > > > > > > * size.
> > > > > > > */
> > > > > > > if (!peer) {
> > > > > > > - return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
> > > > > > > + return VIRTIO_NET_VQ_MAX_SIZE;
> > > > > > > }
> > > > > > >
> > > > > > > switch(peer->info->type) {
> > > > > > > @@ -641,7 +662,7 @@ static int
> > > > > > > virtio_net_max_tx_queue_size(VirtIONet *n)
> > > > > > > case NET_CLIENT_DRIVER_VHOST_VDPA:
> > > > > > > return VIRTQUEUE_MAX_SIZE;
> > > > > > > default:
> > > > > > > - return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
> > > > > > > + return VIRTIO_NET_VQ_MAX_SIZE;
> > > > > > > };
> > > > > > > }
> > > > > > >
> > > > > > > @@ -3450,30 +3471,30 @@ static void
> > > > > > > virtio_net_device_realize(DeviceState *dev, Error **errp)
> > > > > > >
> > > > > > > virtio_net_set_config_size(n, n->host_features);
> > > > > > > virtio_init(vdev, VIRTIO_ID_NET, n->config_size);
> > > > > > > -
> > > > > > > + virtio_net_set_default_queue_size(n);
> > > > > > > /*
> > > > > > > * We set a lower limit on RX queue size to what it always
> > > > > > > was.
> > > > > > > * Guests that want a smaller ring can always resize it
> > > > > > > without
> > > > > > > * help from us (using virtio 1 and up).
> > > > > > > */
> > > > > > > if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE
> > > > > > > ||
> > > > > > > - n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE ||
> > > > > > > + n->net_conf.rx_queue_size > VIRTIO_NET_VQ_MAX_SIZE ||
> > > > > > > !is_power_of_2(n->net_conf.rx_queue_size)) {
> > > > > > > error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "),
> > > > > > > "
> > > > > > > "must be a power of 2 between %d and %d.",
> > > > > > > n->net_conf.rx_queue_size,
> > > > > > > VIRTIO_NET_RX_QUEUE_MIN_SIZE,
> > > > > > > - VIRTQUEUE_MAX_SIZE);
> > > > > > > + VIRTIO_NET_VQ_MAX_SIZE );
> > > > > > > virtio_cleanup(vdev);
> > > > > > > return;
> > > > > > > }
> > > > > > >
> > > > > > > if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE
> > > > > > > ||
> > > > > > > - n->net_conf.tx_queue_size > VIRTQUEUE_MAX_SIZE ||
> > > > > > > + n->net_conf.tx_queue_size > VIRTIO_NET_VQ_MAX_SIZE ||
> > > > > > > !is_power_of_2(n->net_conf.tx_queue_size)) {
> > > > > > > error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "),
> > > > > > > "
> > > > > > > "must be a power of 2 between %d and %d",
> > > > > > > n->net_conf.tx_queue_size,
> > > > > > > VIRTIO_NET_TX_QUEUE_MIN_SIZE,
> > > > > > > - VIRTQUEUE_MAX_SIZE);
> > > > > > > + VIRTIO_NET_VQ_MAX_SIZE);
> > > > > > > virtio_cleanup(vdev);
> > > > > > > return;
> > > > > > > }
> > > > > > > @@ -3751,9 +3772,9 @@ static Property virtio_net_properties[] = {
> > > > > > > DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst,
> > > > > > > TX_BURST),
> > > > > > > DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
> > > > > > > DEFINE_PROP_UINT16("rx_queue_size", VirtIONet,
> > > > > > > net_conf.rx_queue_size,
> > > > > > > - VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE),
> > > > > > > + 0),
> > > > > > > DEFINE_PROP_UINT16("tx_queue_size", VirtIONet,
> > > > > > > net_conf.tx_queue_size,
> > > > > > > - VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE),
> > > > > > > + 0),
> > > > > > > DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0),
> > > > > > > DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet,
> > > > > > > mtu_bypass_backend,
> > > > > > > true),
> > > > > > > diff --git a/hw/virtio/virtio.c b/hw/virtio/virtio.c
> > > > > > > index 5d607aeaa0..ad9dfa20e7 100644
> > > > > > > --- a/hw/virtio/virtio.c
> > > > > > > +++ b/hw/virtio/virtio.c
> > > > > > > @@ -2286,11 +2286,15 @@ void virtio_queue_set_rings(VirtIODevice
> > > > > > > *vdev, int n, hwaddr desc,
> > > > > > >
> > > > > > > void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
> > > > > > > {
> > > > > > > + int vq_max_size = VIRTQUEUE_MAX_SIZE;
> > > > > > > + if (!strcmp(vdev->name, "virtio-net")) {
> > > > > > > + vq_max_size = VIRTIO_NET_VQ_MAX_SIZE;
> > > > > > > + }
> > > > > > > /* Don't allow guest to flip queue between existent and
> > > > > > > * nonexistent states, or to set it to an invalid size.
> > > > > > > */
> > > > > > > if (!!num != !!vdev->vq[n].vring.num ||
> > > > > > > - num > VIRTQUEUE_MAX_SIZE ||
> > > > > > > + num > vq_max_size ||
> > > > > > > num < 0) {
> > > > > > > return;
> > > > > > > }
> > > > > > > @@ -2423,7 +2427,7 @@ VirtQueue *virtio_add_queue(VirtIODevice
> > > > > > > *vdev, int queue_size,
> > > > > > > break;
> > > > > > > }
> > > > > > >
> > > > > > > - if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
> > > > > > > + if (i == VIRTIO_QUEUE_MAX )
> > > > > > > abort();
> > > > > > >
> > > > > > > vdev->vq[i].vring.num = queue_size;
> > > > > > > diff --git a/include/hw/virtio/virtio.h
> > > > > > > b/include/hw/virtio/virtio.h
> > > > > > > index db1c0ddf6b..1f4d2eb5d7 100644
> > > > > > > --- a/include/hw/virtio/virtio.h
> > > > > > > +++ b/include/hw/virtio/virtio.h
> > > > > > > @@ -50,6 +50,7 @@ size_t virtio_feature_get_config_size(const
> > > > > > > VirtIOFeature *features,
> > > > > > > typedef struct VirtQueue VirtQueue;
> > > > > > >
> > > > > > > #define VIRTQUEUE_MAX_SIZE 1024
> > > > > > > +#define VIRTIO_NET_VQ_MAX_SIZE (4096)
> > > > > > >
> > > > > > > typedef struct VirtQueueElement
> > > > > > > {
> > > > > > > --
> > > > > > > 2.27.0
> > > > > > >
> > > >
> >