DPDK vHost User Ports

OVS userspace switching supports vHost user ports as a primary way to interact with guests. For more information on vHost User, refer to the QEMU documentation on same.

Important

To use any DPDK-backed interface, you must ensure your bridge is configured correctly. For more information, refer to DPDK Bridges.

Quick Example

This example demonstrates how to add two dpdkvhostuserclient ports to an existing bridge called br0:

$ ovs-vsctl add-port br0 dpdkvhostclient0 \
    -- set Interface dpdkvhostclient0 type=dpdkvhostuserclient \
       options:vhost-server-path=/tmp/dpdkvhostclient0
$ ovs-vsctl add-port br0 dpdkvhostclient1 \
    -- set Interface dpdkvhostclient1 type=dpdkvhostuserclient \
       options:vhost-server-path=/tmp/dpdkvhostclient1

For the above examples to work, an appropriate server socket must be created at the paths specified (/tmp/dpdkvhostclient0 and /tmp/dpdkvhostclient1). These sockets can be created with QEMU; see the vhost-user client section for details.

vhost-user vs. vhost-user-client

Open vSwitch provides two types of vHost User ports:

  • vhost-user (dpdkvhostuser)

  • vhost-user-client (dpdkvhostuserclient)

vHost User uses a client-server model. The server creates/manages/destroys the vHost User sockets, and the client connects to the server. Depending on which port type you use, dpdkvhostuser or dpdkvhostuserclient, a different configuration of the client-server model is used.

For vhost-user ports, Open vSwitch acts as the server and QEMU the client. This means if OVS dies, all VMs must be restarted. On the other hand, for vhost-user-client ports, OVS acts as the client and QEMU the server. This means OVS can die and be restarted without issue, and it is also possible to restart an instance itself. For this reason, vhost-user-client ports are the preferred type for all known use cases; the only limitation is that vhost-user client mode ports require QEMU version 2.7. Ports of type vhost-user are currently deprecated and will be removed in a future release.

vhost-user

Important

Use of vhost-user ports requires QEMU >= 2.2; vhost-user ports are deprecated.

To use vhost-user ports, you must first add said ports to the switch. DPDK vhost-user ports can have arbitrary names with the exception of forward and backward slashes, which are prohibited. For vhost-user, the port type is dpdkvhostuser:

$ ovs-vsctl add-port br0 vhost-user-1 -- set Interface vhost-user-1 \
    type=dpdkvhostuser

This action creates a socket located at /usr/local/var/run/openvswitch/vhost-user-1, which you must provide to your VM on the QEMU command line.

Note

If you wish for the vhost-user sockets to be created in a sub-directory of /usr/local/var/run/openvswitch, you may specify this directory in the ovsdb like so:

$ ovs-vsctl --no-wait \
    set Open_vSwitch . other_config:vhost-sock-dir=subdir

Once the vhost-user ports have been added to the switch, they must be added to the guest. There are two ways to do this: using QEMU directly, or using libvirt.

Note

IOMMU and Post-copy Live Migration are not supported with vhost-user ports.

Adding vhost-user ports to the guest (QEMU)

To begin, you must attach the vhost-user device sockets to the guest. To do this, you must pass the following parameters to QEMU:

-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=00:00:00:00:00:01,netdev=mynet1

where vhost-user-1 is the name of the vhost-user port added to the switch.

Repeat the above parameters for multiple devices, changing the chardev path and id as necessary. Note that a separate and different chardev path needs to be specified for each vhost-user device. For example you have a second vhost-user port named vhost-user-2, you append your QEMU command line with an additional set of parameters:

-chardev socket,id=char2,path=/usr/local/var/run/openvswitch/vhost-user-2
-netdev type=vhost-user,id=mynet2,chardev=char2,vhostforce
-device virtio-net-pci,mac=00:00:00:00:00:02,netdev=mynet2

In addition, QEMU must allocate the VM’s memory on hugetlbfs. vhost-user ports access a virtio-net device’s virtual rings and packet buffers mapping the VM’s physical memory on hugetlbfs. To enable vhost-user ports to map the VM’s memory into their process address space, pass the following parameters to QEMU:

-object memory-backend-file,id=mem,size=4096M,mem-path=/dev/hugepages,share=on
-numa node,memdev=mem -mem-prealloc

Finally, you may wish to enable multiqueue support. This is optional but, should you wish to enable it, run:

-chardev socket,id=char2,path=/usr/local/var/run/openvswitch/vhost-user-2
-netdev type=vhost-user,id=mynet2,chardev=char2,vhostforce,queues=$q
-device virtio-net-pci,mac=00:00:00:00:00:02,netdev=mynet2,mq=on,vectors=$v

where:

$q

The number of queues

$v

The number of vectors, which is $q * 2 + 2

The vhost-user interface will be automatically reconfigured with required number of Rx and Tx queues after connection of virtio device. Manual configuration of n_rxq is not supported because OVS will work properly only if n_rxq will match number of queues configured in QEMU.

A least two PMDs should be configured for the vswitch when using multiqueue. Using a single PMD will cause traffic to be enqueued to the same vhost queue rather than being distributed among different vhost queues for a vhost-user interface.

If traffic destined for a VM configured with multiqueue arrives to the vswitch via a physical DPDK port, then the number of Rx queues should also be set to at least two for that physical DPDK port. This is required to increase the probability that a different PMD will handle the multiqueue transmission to the guest using a different vhost queue.

If one wishes to use multiple queues for an interface in the guest, the driver in the guest operating system must be configured to do so. It is recommended that the number of queues configured be equal to $q.

For example, this can be done for the Linux kernel virtio-net driver with:

$ ethtool -L <DEV> combined <$q>

where:

-L

Changes the numbers of channels of the specified network device

combined

Changes the number of multi-purpose channels.

Adding vhost-user ports to the guest (libvirt)

To begin, you must change the user and group that qemu runs under, and restart libvirtd.

  • In /etc/libvirt/qemu.conf add/edit the following lines:

    user = "root"
    group = "root"
    
  • Finally, restart the libvirtd process, For example, on Fedora:

    $ systemctl restart libvirtd.service
    

Once complete, instantiate the VM. A sample XML configuration file is provided at the end of this file. Save this file, then create a VM using this file:

$ virsh create demovm.xml

Once created, you can connect to the guest console:

$ virsh console demovm

The demovm xml configuration is aimed at achieving out of box performance on VM. These enhancements include:

  • The vcpus are pinned to the cores of the CPU socket 0 using vcpupin.

  • Configure NUMA cell and memory shared using memAccess='shared'.

  • Disable mrg_rxbuf='off'

Refer to the libvirt documentation for more information.

vhost-user-client

Important

Use of vhost-user-client ports requires QEMU >= 2.7

To use vhost-user-client ports, you must first add said ports to the switch. Like DPDK vhost-user ports, DPDK vhost-user-client ports can have mostly arbitrary names. However, the name given to the port does not govern the name of the socket device. Instead, this must be configured by the user by way of a vhost-server-path option. For vhost-user-client, the port type is dpdkvhostuserclient:

$ VHOST_USER_SOCKET_PATH=/path/to/socket
$ ovs-vsctl add-port br0 vhost-client-1 \
    -- set Interface vhost-client-1 type=dpdkvhostuserclient \
         options:vhost-server-path=$VHOST_USER_SOCKET_PATH

Once the vhost-user-client ports have been added to the switch, they must be added to the guest. Like vhost-user ports, there are two ways to do this: using QEMU directly, or using libvirt. Only the QEMU case is covered here.

Adding vhost-user-client ports to the guest (QEMU)

Attach the vhost-user device sockets to the guest. To do this, you must pass the following parameters to QEMU:

-chardev socket,id=char1,path=$VHOST_USER_SOCKET_PATH,server
-netdev type=vhost-user,id=mynet1,chardev=char1,vhostforce
-device virtio-net-pci,mac=00:00:00:00:00:01,netdev=mynet1

where vhost-user-1 is the name of the vhost-user port added to the switch.

If the corresponding dpdkvhostuserclient port has not yet been configured in OVS with vhost-server-path=/path/to/socket, QEMU will print a log similar to the following:

QEMU waiting for connection on: disconnected:unix:/path/to/socket,server

QEMU will wait until the port is created successfully in OVS to boot the VM. One benefit of using this mode is the ability for vHost ports to ‘reconnect’ in event of the switch crashing or being brought down. Once it is brought back up, the vHost ports will reconnect automatically and normal service will resume.

vhost-user-client IOMMU Support

vhost IOMMU is a feature which restricts the vhost memory that a virtio device can access, and as such is useful in deployments in which security is a concern.

IOMMU support may be enabled via a global config value, vhost-iommu-support. Setting this to true enables vhost IOMMU support for all vhost ports when/where available:

$ ovs-vsctl set Open_vSwitch . other_config:vhost-iommu-support=true

The default value is false.

Important

Changing this value requires restarting the daemon.

Important

Enabling the IOMMU feature also enables the vhost user reply-ack protocol; this is known to work on QEMU v2.10.0, but is buggy on older versions (2.7.0 - 2.9.0, inclusive). Consequently, the IOMMU feature is disabled by default (and should remain so if using the aforementioned versions of QEMU). Starting with QEMU v2.9.1, vhost-iommu-support can safely be enabled, even without having an IOMMU device, with no performance penalty.

vhost-user-client Post-copy Live Migration Support (experimental)

Post-copy migration is the migration mode where the destination CPUs are started before all the memory has been transferred. The main advantage is the predictable migration time. Mostly used as a second phase after the normal ‘pre-copy’ migration in case it takes too long to converge.

More information can be found in QEMU docs.

Post-copy support may be enabled via a global config value vhost-postcopy-support. Setting this to true enables Post-copy support for all vhost-user-client ports:

$ ovs-vsctl set Open_vSwitch . other_config:vhost-postcopy-support=true

The default value is false.

Important

Changing this value requires restarting the daemon.

Important

DPDK Post-copy migration mode uses userfaultfd syscall to communicate with the kernel about page fault handling and uses shared memory based on huge pages. So destination host linux kernel should support userfaultfd over shared hugetlbfs. This feature only introduced in kernel upstream version 4.11.

Post-copy feature supported in DPDK since 18.11.0 version and in QEMU since 2.12.0 version. But it’s suggested to use QEMU >= 3.0.1 because migration recovery was fixed for post-copy in 3.0 and few additional bug fixes (like userfaulfd leak) was released in 3.0.1.

DPDK Post-copy feature requires avoiding to populate the guest memory (application must not call mlock* syscall). So enabling mlockall is incompatible with post-copy feature.

Note that during migration of vhost-user device, PMD threads hang for the time of faulted pages download from source host. Transferring 1GB hugepage across a 10Gbps link possibly unacceptably slow. So recommended hugepage size is 2MB.

vhost-user-client tx retries config

For vhost-user-client interfaces, the max amount of retries can be changed from the default 8 by setting tx-retries-max.

The minimum is 0 which means there will be no retries and if any packets in each batch cannot be sent immediately they will be dropped. The maximum is 32, which would mean that after the first packet(s) in the batch was sent there could be a maximum of 32 more retries.

Retries can help with avoiding packet loss when temporarily unable to send to a vhost interface because the virtqueue is full. However, spending more time retrying to send to one interface, will reduce the time available for rx/tx and processing packets on other interfaces, so some tuning may be required for best performance.

Tx retries max can be set for vhost-user-client ports:

$ ovs-vsctl set Interface vhost-client-1 options:tx-retries-max=0

Note

Configurable vhost tx retries are not supported with vhost-user ports.

DPDK in the Guest

The DPDK testpmd application can be run in guest VMs for high speed packet forwarding between vhostuser ports. DPDK and testpmd application has to be compiled on the guest VM. Below are the steps for setting up the testpmd application in the VM.

Note

Support for DPDK in the guest requires QEMU >= 2.2

To begin, instantiate a guest as described in vhost-user or vhost-user-client. Once started, connect to the VM, download the DPDK sources to VM and build DPDK as described in Installing.

Setup huge pages and DPDK devices using UIO:

$ sysctl vm.nr_hugepages=1024
$ mkdir -p /dev/hugepages
$ mount -t hugetlbfs hugetlbfs /dev/hugepages  # only if not already mounted
$ modprobe uio
$ insmod $DPDK_BUILD/kmod/igb_uio.ko
$ $DPDK_DIR/usertools/dpdk-devbind.py --status
$ $DPDK_DIR/usertools/dpdk-devbind.py -b igb_uio 00:03.0 00:04.0

Note

vhost ports pci ids can be retrieved using:

lspci | grep Ethernet

Finally, start the application:

# TODO

Sample XML

<domain type='kvm'>
  <name>demovm</name>
  <uuid>4a9b3f53-fa2a-47f3-a757-dd87720d9d1d</uuid>
  <memory unit='KiB'>4194304</memory>
  <currentMemory unit='KiB'>4194304</currentMemory>
  <memoryBacking>
    <hugepages>
      <page size='2' unit='M' nodeset='0'/>
    </hugepages>
  </memoryBacking>
  <vcpu placement='static'>2</vcpu>
  <cputune>
    <shares>4096</shares>
    <vcpupin vcpu='0' cpuset='4'/>
    <vcpupin vcpu='1' cpuset='5'/>
    <emulatorpin cpuset='4,5'/>
  </cputune>
  <os>
    <type arch='x86_64' machine='pc'>hvm</type>
    <boot dev='hd'/>
  </os>
  <features>
    <acpi/>
    <apic/>
  </features>
  <cpu mode='host-model'>
    <model fallback='allow'/>
    <topology sockets='2' cores='1' threads='1'/>
    <numa>
      <cell id='0' cpus='0-1' memory='4194304' unit='KiB' memAccess='shared'/>
    </numa>
  </cpu>
  <on_poweroff>destroy</on_poweroff>
  <on_reboot>restart</on_reboot>
  <on_crash>destroy</on_crash>
  <devices>
    <emulator>/usr/bin/qemu-system-x86_64</emulator>
    <disk type='file' device='disk'>
      <driver name='qemu' type='qcow2' cache='none'/>
      <source file='/root/CentOS7_x86_64.qcow2'/>
      <target dev='vda' bus='virtio'/>
    </disk>
    <interface type='vhostuser'>
      <mac address='00:00:00:00:00:01'/>
      <source type='unix' path='/usr/local/var/run/openvswitch/dpdkvhostuser0' mode='client'/>
       <model type='virtio'/>
      <driver queues='2'>
        <host mrg_rxbuf='on'/>
      </driver>
    </interface>
    <interface type='vhostuser'>
      <mac address='00:00:00:00:00:02'/>
      <source type='unix' path='/usr/local/var/run/openvswitch/dpdkvhostuser1' mode='client'/>
      <model type='virtio'/>
      <driver queues='2'>
        <host mrg_rxbuf='on'/>
      </driver>
    </interface>
    <serial type='pty'>
      <target port='0'/>
    </serial>
    <console type='pty'>
      <target type='serial' port='0'/>
    </console>
  </devices>
</domain>

Jumbo Frames

DPDK vHost User ports can be configured to use Jumbo Frames. For more information, refer to Jumbo Frames.

vhost tx retries

When sending a batch of packets to a vhost-user or vhost-user-client interface, it may happen that some but not all of the packets in the batch are able to be sent to the guest. This is often because there is not enough free descriptors in the virtqueue for all the packets in the batch to be sent. In this case there will be a retry, with a default maximum of 8 occurring. If at any time no packets can be sent, it may mean the guest is not accepting packets, so there are no (more) retries.

For information about configuring the maximum amount of tx retries for vhost-user-client interfaces see vhost-user-client tx retries config.

Note

Maximum vhost tx batch size is defined by NETDEV_MAX_BURST, and is currently as 32.

Tx Retries may be reduced or even avoided by some external configuration, such as increasing the virtqueue size through the rx_queue_size parameter introduced in QEMU 2.7.0 / libvirt 2.3.0:

<interface type='vhostuser'>
    <mac address='56:48:4f:53:54:01'/>
    <source type='unix' path='/tmp/dpdkvhostclient0' mode='server'/>
    <model type='virtio'/>
    <driver name='vhost' rx_queue_size='1024' tx_queue_size='1024'/>
    <address type='pci' domain='0x0000' bus='0x00' slot='0x10' function='0x0'/>
</interface>

The guest application will also need to provide enough descriptors. For example with testpmd the command line argument can be used:

--rxd=1024 --txd=1024

The guest should also have sufficient cores dedicated for consuming and processing packets at the required rate.

The amount of Tx retries on a vhost-user or vhost-user-client interface can be shown with:

$ ovs-vsctl get Interface dpdkvhostclient0 statistics:ovs_tx_retries

Further information can be found in the DPDK documentation