Re: [virtio-dev] Re: [PATCH v4] virtio-net: enable configurable tx queue size

Re: [virtio-dev] Re: [PATCH v4] virtio-net: enable configurable tx queue size

[Michael S. Tsirkin]

On Tue, Jul 04, 2017 at 07:03:51PM +0800, Wei Wang wrote:
> On 07/04/2017 03:18 AM, Michael S. Tsirkin wrote:
> > On Wed, Jun 28, 2017 at 10:37:59AM +0800, Wei Wang wrote:
> > > This patch enables the virtio-net tx queue size to be configurable
> > > between 256 (the default queue size) and 1024 by the user when the
> > > vhost-user backend is used.
> > > 
> > > Currently, the maximum tx queue size for other backends is 512 due
> > > to the following limitations:
> > > - QEMU backend: the QEMU backend implementation in some cases may
> > > send 1024+1 iovs to writev.
> > > - Vhost_net backend: there are possibilities that the guest sends
> > > a vring_desc of memory which crosses a MemoryRegion thereby
> > > generating more than 1024 iovs after translation from guest-physical
> > > address in the backend.
> > > 
> > > Signed-off-by: Wei Wang <wei.w.wang@intel.com>
> > I was going to apply this, but run into a host of issues:
> > 
> > This segfaults:
> > $ ./x86_64-softmmu/qemu-system-x86_64 -device virtio-net,tx_queue_size=1024
> > Segmentation fault (core dumped)
> > 
> > I tried to tweak this code a bit to avoid the crash, and I run into a further issue:
> > $ ./x86_64-softmmu/qemu-system-x86_64 -device virtio-net,tx_queue_size=1024
> > Bad ram offset aa49002
> > Aborted (core dumped)
> > 
> > the second issue is especially concerning.
> > 
> 
> AFAIK, all the virtio-net backends require "-netdev". I'm wondering if there
> is any case that virtio-net can work without a "-netdev" created in QEMU?

Of course. Old style -net with vlans still work.

> If not, would it be better if we just stop the device creation at the
> beginning of
> virtio_net_device_realize() if "-netdev" is not given (i.e.
> !n->nic_conf.peers.ncs[0])?
> 
> Best,
> Wei

That will break a ton of scripts without any real benefit
to users.


-- 
MST

---------------------------------------------------------------------
To unsubscribe, e-mail: virtio-dev-unsubscribe@lists.oasis-open.org
For additional commands, e-mail: virtio-dev-help@lists.oasis-open.org

[virtio-dev] Re: [PATCH v4] virtio-net: enable configurable tx queue size

[Michael S. Tsirkin]

On Wed, Jun 28, 2017 at 10:37:59AM +0800, Wei Wang wrote:
> This patch enables the virtio-net tx queue size to be configurable
> between 256 (the default queue size) and 1024 by the user when the
> vhost-user backend is used.
> 
> Currently, the maximum tx queue size for other backends is 512 due
> to the following limitations:
> - QEMU backend: the QEMU backend implementation in some cases may
> send 1024+1 iovs to writev.
> - Vhost_net backend: there are possibilities that the guest sends
> a vring_desc of memory which crosses a MemoryRegion thereby
> generating more than 1024 iovs after translation from guest-physical
> address in the backend.
> 
> Signed-off-by: Wei Wang <wei.w.wang@intel.com>

Could you pls add a bit info about how this was tested?
Was any special setup for dpdk necessary?

Thanks!

> ---
>  hw/net/virtio-net.c            | 32 ++++++++++++++++++++++++++++++--
>  include/hw/virtio/virtio-net.h |  1 +
>  2 files changed, 31 insertions(+), 2 deletions(-)
> 
> diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
> index 91eddaf..a1fc0db 100644
> --- a/hw/net/virtio-net.c
> +++ b/hw/net/virtio-net.c
> @@ -34,8 +34,11 @@
>  
>  /* previously fixed value */
>  #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
> +#define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256
> +
>  /* for now, only allow larger queues; 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
>  
>  /*
>   * Calculate the number of bytes up to and including the given 'field' of
> @@ -1508,15 +1511,18 @@ static void virtio_net_add_queue(VirtIONet *n, int index)
>  
>      n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size,
>                                             virtio_net_handle_rx);
> +
>      if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
>          n->vqs[index].tx_vq =
> -            virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
> +            virtio_add_queue(vdev, n->net_conf.tx_queue_size,
> +                             virtio_net_handle_tx_timer);
>          n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
>                                                virtio_net_tx_timer,
>                                                &n->vqs[index]);
>      } else {
>          n->vqs[index].tx_vq =
> -            virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
> +            virtio_add_queue(vdev, n->net_conf.tx_queue_size,
> +                             virtio_net_handle_tx_bh);
>          n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]);
>      }
>  
> @@ -1927,6 +1933,17 @@ static void virtio_net_device_realize(DeviceState *dev, Error **errp)
>          return;
>      }
>  
> +    if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE ||
> +        n->net_conf.tx_queue_size > VIRTQUEUE_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_cleanup(vdev);
> +        return;
> +    }
> +
>      n->max_queues = MAX(n->nic_conf.peers.queues, 1);
>      if (n->max_queues * 2 + 1 > VIRTIO_QUEUE_MAX) {
>          error_setg(errp, "Invalid number of queues (= %" PRIu32 "), "
> @@ -1947,6 +1964,15 @@ static void virtio_net_device_realize(DeviceState *dev, Error **errp)
>          error_report("Defaulting to \"bh\"");
>      }
>  
> +    /*
> +     * Currently, backends other than vhost-user don't support 1024 queue
> +     * size.
> +     */
> +    if (n->net_conf.tx_queue_size == VIRTQUEUE_MAX_SIZE &&
> +        n->nic_conf.peers.ncs[0]->info->type != NET_CLIENT_DRIVER_VHOST_USER) {
> +        n->net_conf.tx_queue_size = VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
> +    }
> +
>      for (i = 0; i < n->max_queues; i++) {
>          virtio_net_add_queue(n, i);
>      }
> @@ -2106,6 +2132,8 @@ static Property virtio_net_properties[] = {
>      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),
> +    DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size,
> +                       VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE),
>      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/include/hw/virtio/virtio-net.h b/include/hw/virtio/virtio-net.h
> index 602b486..b81b6a4 100644
> --- a/include/hw/virtio/virtio-net.h
> +++ b/include/hw/virtio/virtio-net.h
> @@ -36,6 +36,7 @@ typedef struct virtio_net_conf
>      int32_t txburst;
>      char *tx;
>      uint16_t rx_queue_size;
> +    uint16_t tx_queue_size;
>      uint16_t mtu;
>  } virtio_net_conf;
>  
> -- 
> 2.7.4

---------------------------------------------------------------------
To unsubscribe, e-mail: virtio-dev-unsubscribe@lists.oasis-open.org
For additional commands, e-mail: virtio-dev-help@lists.oasis-open.org

[virtio-dev] RE: [PATCH v4] virtio-net: enable configurable tx queue size

[Wang, Wei W]

On Thursday, July 6, 2017 9:49 PM, Michael S. Tsirkin wrote:
> On Wed, Jun 28, 2017 at 10:37:59AM +0800, Wei Wang wrote:
> > This patch enables the virtio-net tx queue size to be configurable
> > between 256 (the default queue size) and 1024 by the user when the
> > vhost-user backend is used.
> >
> > Currently, the maximum tx queue size for other backends is 512 due to
> > the following limitations:
> > - QEMU backend: the QEMU backend implementation in some cases may send
> > 1024+1 iovs to writev.
> > - Vhost_net backend: there are possibilities that the guest sends a
> > vring_desc of memory which crosses a MemoryRegion thereby generating
> > more than 1024 iovs after translation from guest-physical address in
> > the backend.
> >
> > Signed-off-by: Wei Wang <wei.w.wang@intel.com>
> 
> Could you pls add a bit info about how this was tested?
> Was any special setup for dpdk necessary?

Yes, I used the vhost-user implementation in DPDK. So, on the host, I have 
the legacy ovs-dpdk setup ready (I'm using dpdk-stable-16.11.1 and openvswitch-2.6.1,
the setup steps can be found inside the source code directory).

When booting the guest, I have the following QEMU commands:
-chardev socket,id=char1,path=/usr/local/var/run/openvswitch/vhost-user-1
-netdev type=vhost-user,id=mynet1,chardev=char1,vhostforce
-device virtio-net-pci,mac=52:54:00:00:00:01,netdev=mynet1,tx_queue_size=1024

To check the guest tx queue size, I simply added a printk() at the end of virtnet_probe()
to print out vi->sq->vq->num_free, which initially equals to the queue size.

Then, I did Ping and netperf tests to transmit packets between VMs, which worked fine.

If the related configuration support to Libvirt is ready, I think we can get the customer
to try in their test environment, too.

Best,
Wei






---------------------------------------------------------------------
To unsubscribe, e-mail: virtio-dev-unsubscribe@lists.oasis-open.org
For additional commands, e-mail: virtio-dev-help@lists.oasis-open.org