<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE html>
<html xmlns="http://www.w3.org/1999/xhtml">
  <body>
    <h1>Host device management</h1>

    <p>
      Libvirt provides management of both physical and virtual host devices
      (historically also referred to as node devices) like USB, PCI, SCSI, and
      network devices. This also includes various virtualization capabilities
      which the aforementioned devices provide for utilization, for example
      SR-IOV, NPIV, MDEV, DRM, etc.
    </p>

    <p>
      The node device driver provides means to list and show details about host
      devices (<code>virsh nodedev-list</code>, <code>virsh nodedev-info</code>,
      and <code>virsh nodedev-dumpxml</code>), which are generic and can be used
      with all devices.  It also provides the means to manage virtual devices.
      Persistently-defined virtual devices are only supported for mediated
      devices, while transient devices are supported by both mediated devices
      and NPIV (<a href="https://wiki.libvirt.org/page/NPIV_in_libvirt">more
      info about NPIV)</a>).
    </p>
    <p>
      Persistent virtual devices are managed with
      <code>virsh nodedev-define</code> and <code>virsh nodedev-undefine</code>.
      Persistent devices can be configured to start manually or automatically
      using <code>virsh nodedev-autostart</code>. Inactive devices can be made
      active with <code>virsh nodedev-start</code>.
    </p>
    <p>
      Transient virtual devices are started and stopped with the commands
      <code>virsh nodedev-create</code> and <code>virsh nodedev-destroy</code>.
    </p>
    <p>
      Devices on the host system are arranged in a tree-like hierarchy, with
      the root node being called <code>computer</code>. The node device driver
      supports udev backend (HAL backend was removed in <code>6.8.0</code>).
    </p>

    <p>
      Details of the XML format of a host device can be found <a
      href="formatnode.html">here</a>. Of particular interest is the
      <code>capability</code> element, which describes features supported by
      the device. Some specific device types are addressed in more detail
      below.
    </p>
    <h2>Basic structure of a node device</h2>
    <pre>
&lt;device&gt;
  &lt;name&gt;pci_0000_00_17_0&lt;/name&gt;
  &lt;path&gt;/sys/devices/pci0000:00/0000:00:17.0&lt;/path&gt;
  &lt;parent&gt;computer&lt;/parent&gt;
  &lt;driver&gt;
    &lt;name&gt;ahci&lt;/name&gt;
  &lt;/driver&gt;
  &lt;capability type='pci'&gt;
...
  &lt;/capability&gt;
&lt;/device&gt;</pre>

    <ul id="toc"/>

    <h2><a id="PCI">PCI host devices</a></h2>
    <dl>
      <dt><code>capability</code></dt>
      <dd>
        When used as top level element, the supported values for the
        <code>type</code> attribute are <code>pci</code> and
        <code>phys_function</code> (see <a href="#SRIOVCap">SR-IOV below</a>).
      </dd>
    </dl>
    <pre>
&lt;device&gt;
  &lt;name&gt;pci_0000_04_00_1&lt;/name&gt;
  &lt;path&gt;/sys/devices/pci0000:00/0000:00:06.0/0000:04:00.1&lt;/path&gt;
  &lt;parent&gt;pci_0000_00_06_0&lt;/parent&gt;
  &lt;driver&gt;
    &lt;name&gt;igb&lt;/name&gt;
  &lt;/driver&gt;
  &lt;capability type='pci'&gt;
    &lt;domain&gt;0&lt;/domain&gt;
    &lt;bus&gt;4&lt;/bus&gt;
    &lt;slot&gt;0&lt;/slot&gt;
    &lt;function&gt;1&lt;/function&gt;
    &lt;product id='0x10c9'&gt;82576 Gigabit Network Connection&lt;/product&gt;
    &lt;vendor id='0x8086'&gt;Intel Corporation&lt;/vendor&gt;
    &lt;iommuGroup number='15'&gt;
      &lt;address domain='0x0000' bus='0x04' slot='0x00' function='0x1'/&gt;
    &lt;/iommuGroup&gt;
    &lt;numa node='0'/&gt;
    &lt;pci-express&gt;
      &lt;link validity='cap' port='1' speed='2.5' width='2'/&gt;
      &lt;link validity='sta' speed='2.5' width='2'/&gt;
    &lt;/pci-express&gt;
  &lt;/capability&gt;
&lt;/device&gt;</pre>

    <p>
      The XML format for a PCI device stays the same for any further
      capabilities it supports, a single nested <code>&lt;capability&gt;</code>
      element will be included for each capability the device supports.
    </p>

    <h3><a id="SRIOVCap">SR-IOV capability</a></h3>
    <p>
      Single root input/output virtualization (SR-IOV) allows sharing of the
      PCIe resources by multiple virtual environments. That is achieved by
      slicing up a single full-featured physical resource called physical
      function (PF) into multiple devices called virtual functions (VFs) sharing
      their configuration with the underlying PF. Despite the SR-IOV
      specification, the amount of VFs that can be created on a PF varies among
      manufacturers.
    </p>

    <p>
      Suppose the NIC <a href="#PCI">above</a> was also SR-IOV capable, it would
      also include a nested
      <code>&lt;capability&gt;</code> element enumerating all virtual
      functions available on the physical device (physical port) like in the
      example below.
    </p>

    <pre>
&lt;capability type='pci'&gt;
...
  &lt;capability type='virt_functions' maxCount='7'&gt;
    &lt;address domain='0x0000' bus='0x04' slot='0x10' function='0x1'/&gt;
    &lt;address domain='0x0000' bus='0x04' slot='0x10' function='0x3'/&gt;
    &lt;address domain='0x0000' bus='0x04' slot='0x10' function='0x5'/&gt;
    &lt;address domain='0x0000' bus='0x04' slot='0x10' function='0x7'/&gt;
    &lt;address domain='0x0000' bus='0x04' slot='0x11' function='0x1'/&gt;
    &lt;address domain='0x0000' bus='0x04' slot='0x11' function='0x3'/&gt;
    &lt;address domain='0x0000' bus='0x04' slot='0x11' function='0x5'/&gt;
  &lt;/capability&gt;
...
&lt;/capability&gt;</pre>
    <p>
      A SR-IOV child device on the other hand, would then report its top level
      capability type as a <code>phys_function</code> instead:
    </p>

    <pre>
&lt;device&gt;
...
  &lt;capability type='phys_function'&gt;
    &lt;address domain='0x0000' bus='0x04' slot='0x00' function='0x0'/&gt;
  &lt;/capability&gt;
...
&lt;/device&gt;</pre>

    <h3><a id="MDEVCap">MDEV capability</a></h3>
    <p>
      A device capable of creating mediated devices will include a nested
      capability <code>mdev_types</code> which enumerates all supported mdev
      types on the physical device, along with the type attributes available
      through sysfs. A detailed description of the XML format for the
      <code>mdev_types</code> capability can be found
      <a href="formatnode.html#MDEVTypesCap">here</a>.
    </p>
    <p>
      The following example shows how we might represent an NVIDIA GPU device
      that supports mediated devices. See below for <a href="#MDEV">more
      information about mediated devices</a>.
    </p>

<pre>
&lt;device&gt;
...
  &lt;driver&gt;
    &lt;name&gt;nvidia&lt;/name&gt;
  &lt;/driver&gt;
  &lt;capability type='pci'&gt;
...
    &lt;capability type='mdev_types'&gt;
      &lt;type id='nvidia-11'&gt;
        &lt;name&gt;GRID M60-0B&lt;/name&gt;
        &lt;deviceAPI&gt;vfio-pci&lt;/deviceAPI&gt;
        &lt;availableInstances&gt;16&lt;/availableInstances&gt;
      &lt;/type&gt;
      &lt;!-- Here would come the rest of the available mdev types --&gt;
    &lt;/capability&gt;
...
  &lt;/capability&gt;
&lt;/device&gt;</pre>

    <h3><a id="VPDCap">VPD capability</a></h3>
    <p>
      A device that exposes a PCI/PCIe VPD capability will include a nested
      capability <code>vpd</code> which presents data stored in the Vital Product
      Data (VPD). VPD provides a device name and a number of other standard-defined
      read-only fields (change level, manufacture id, part number, serial number) and
      vendor-specific read-only fields. Additionally, if a device supports it,
      read-write fields (asset tag, vendor-specific fields or system fields) may
      also be present. The VPD capability is optional for PCI/PCIe devices and the
      set of exposed fields may vary depending on a device. The XML format follows
      the binary format described in "I.3. VPD Definitions" in PCI Local Bus (2.2+)
      and the identical format in PCIe 4.0+. At the time of writing, the support for
      exposing this capability is only present on Linux-based systems (kernel version
      v2.6.26 is the first one to expose VPD via sysfs which Libvirt relies on).
      Reading the VPD contents requires root privileges, therefore,
      <code>virsh nodedev-dumpxml</code> must be executed accordingly.
      A description of the XML format for the <code>vpd</code> capability can
      be found <a href="formatnode.html#VPDCap">here</a>.
    </p>
    <p>
      The following example shows a VPD representation for a device that exposes the
      VPD capability with read-only and read-write fields. Among other things,
      the VPD of this particular device includes a unique board serial number.
    </p>
<pre>
&lt;device&gt;
  &lt;name&gt;pci_0000_42_00_0&lt;/name&gt;
  &lt;capability type=&apos;pci&apos;&gt;
    &lt;class&gt;0x020000&lt;/class&gt;
    &lt;domain&gt;0&lt;/domain&gt;
    &lt;bus&gt;66&lt;/bus&gt;
    &lt;slot&gt;0&lt;/slot&gt;
    &lt;function&gt;0&lt;/function&gt;
    &lt;product id=&apos;0xa2d6&apos;&gt;MT42822 BlueField-2 integrated ConnectX-6 Dx network controller&lt;/product&gt;
    &lt;vendor id=&apos;0x15b3&apos;&gt;Mellanox Technologies&lt;/vendor&gt;
    &lt;capability type=&apos;virt_functions&apos; maxCount=&apos;16&apos;/&gt;
    &lt;capability type=&apos;vpd&apos;&gt;
      &lt;name&gt;BlueField-2 DPU 25GbE Dual-Port SFP56, Crypto Enabled, 16GB on-board DDR, 1GbE OOB management, Tall Bracket&lt;/name&gt;
      &lt;fields access=&apos;readonly&apos;&gt;
        &lt;change_level&gt;B1&lt;/change_level&gt;
        &lt;manufacture_id&gt;foobar&lt;/manufacture_id&gt;
        &lt;part_number&gt;MBF2H332A-AEEOT&lt;/part_number&gt;
        &lt;serial_number&gt;MT2113X00000&lt;/serial_number&gt;
        &lt;vendor_field index=&apos;0&apos;&gt;PCIeGen4 x8&lt;/vendor_field&gt;
        &lt;vendor_field index=&apos;2&apos;&gt;MBF2H332A-AEEOT&lt;/vendor_field&gt;
        &lt;vendor_field index=&apos;3&apos;&gt;3c53d07eec484d8aab34dabd24fe575aa&lt;/vendor_field&gt;
        &lt;vendor_field index=&apos;A&apos;&gt;MLX:MN=MLNX:CSKU=V2:UUID=V3:PCI=V0:MODL=BF2H332A&lt;/vendor_field&gt;
      &lt;/fields&gt;
      &lt;fields access=&apos;readwrite&apos;&gt;
        &lt;asset_tag&gt;fooasset&lt;/asset_tag&gt;
        &lt;vendor_field index=&apos;0&apos;&gt;vendorfield0&lt;/vendor_field&gt;
        &lt;vendor_field index=&apos;2&apos;&gt;vendorfield2&lt;/vendor_field&gt;
        &lt;vendor_field index=&apos;A&apos;&gt;vendorfieldA&lt;/vendor_field&gt;
        &lt;system_field index=&apos;B&apos;&gt;systemfieldB&lt;/system_field&gt;
        &lt;system_field index=&apos;0&apos;&gt;systemfield0&lt;/system_field&gt;
      &lt;/fields&gt;
    &lt;/capability&gt;
    &lt;iommuGroup number=&apos;65&apos;&gt;
      &lt;address domain=&apos;0x0000&apos; bus=&apos;0x42&apos; slot=&apos;0x00&apos; function=&apos;0x0&apos;/&gt;
    &lt;/iommuGroup&gt;
    &lt;numa node=&apos;0&apos;/&gt;
    &lt;pci-express&gt;
      &lt;link validity=&apos;cap&apos; port=&apos;0&apos; speed=&apos;16&apos; width=&apos;8&apos;/&gt;
      &lt;link validity=&apos;sta&apos; speed=&apos;8&apos; width=&apos;8&apos;/&gt;
    &lt;/pci-express&gt;
  &lt;/capability&gt;
&lt;/device&gt;
</pre>

    <h2><a id="MDEV">Mediated devices (MDEVs)</a></h2>
    <p>
      Mediated devices (<span class="since">Since 3.2.0</span>) are software
      devices defining resource allocation on the backing physical device which
      in turn allows the parent physical device's resources to be divided into
      several mediated devices, thus sharing the physical device's performance
      among multiple guests. Unlike SR-IOV however, where a PCIe device appears
      as multiple separate PCIe devices on the host's PCI bus, mediated devices
      only appear on the mdev virtual bus. Therefore, no detach/reattach
      procedure from/to the host driver procedure is involved even though
      mediated devices are used in a direct device assignment manner.  A
      detailed description of the XML format for the <code>mdev</code>
      capability can be found <a href="formatnode.html#mdev">here</a>.
    </p>

    <h3>Example of a mediated device</h3>
    <pre>
&lt;device&gt;
  &lt;name&gt;mdev_4b20d080_1b54_4048_85b3_a6a62d165c01&lt;/name&gt;
  &lt;path&gt;/sys/devices/pci0000:00/0000:00:02.0/4b20d080-1b54-4048-85b3-a6a62d165c01&lt;/path&gt;
  &lt;parent&gt;pci_0000_06_00_0&lt;/parent&gt;
  &lt;driver&gt;
    &lt;name&gt;vfio_mdev&lt;/name&gt;
  &lt;/driver&gt;
  &lt;capability type='mdev'&gt;
    &lt;type id='nvidia-11'/&gt;
    &lt;uuid&gt;4b20d080-1b54-4048-85b3-a6a62d165c01&lt;/uuid&gt;
    &lt;iommuGroup number='12'/&gt;
  &lt;/capability&gt;
&lt;/device&gt;</pre>

    <p>
      The support of mediated device's framework in libvirt's node device driver
      covers the following features:
    </p>

    <ul>
      <li>
        list available mediated devices on the host
        (<span class="since">Since 3.4.0</span>)
      </li>
      <li>
        display device details
        (<span class="since">Since 3.4.0</span>)
      </li>
      <li>
        create transient mediated devices
        (<span class="since">Since 6.5.0</span>)
      </li>
      <li>
        define persistent mediated devices
        (<span class="since">Since 7.3.0</span>)
      </li>
    </ul>

    <p>
      Because mediated devices are instantiated from vendor specific templates,
      simply called 'types', information describing these types is contained
      within the parent device's capabilities (see the example in <a
      href="#PCI">PCI host devices</a>). To list all devices capable of
      creating mediated devices, the following command can be used.
    </p>
    <pre>$ virsh nodedev-list --cap mdev_types</pre>

    <p>
      To see the supported mediated device types on a specific physical device
      use the following:
    </p>

    <pre>$ virsh nodedev-dumpxml &lt;device&gt;</pre>

    <p>
      Before creating a mediated device, unbind the device from the respective
      device driver, eg. subchannel I/O driver for a CCW device. Then bind the
      device to the respective VFIO driver. For a CCW device, also unbind the
      corresponding subchannel of the CCW device from the subchannel I/O driver
      and then bind the subchannel (instead of the CCW device) to the vfio_ccw
      driver. The below example shows the unbinding and binding steps for a CCW
      device.
    </p>

    <pre>
device="0.0.1234"
subchannel="0.0.0123"
echo $device &gt; /sys/bus/ccw/devices/$device/driver/unbind
echo $subchannel &gt; /sys/bus/css/devices/$subchannel/driver/unbind
echo $subchannel &gt; /sys/bus/css/drivers/vfio_ccw/bind
    </pre>

    <p>
      To instantiate a transient mediated device, create an XML file representing the
      device. See above for information about the mediated device xml format.
    </p>

    <pre>$ virsh nodedev-create &lt;xml-file&gt;
Node device '&lt;device-name&gt;' created from '&lt;xml-file&gt;'</pre>

    <p>
      If you would like to persistently define the device so that it will be
      maintained across host reboots, use <code>virsh nodedev-define</code>
      instead of <code>nodedev-create</code>:
    </p>

    <pre>$ virsh nodedev-define &lt;xml-file&gt;
Node device '&lt;device-name&gt;' defined from '&lt;xml-file&gt;'</pre>

    <p>
      To start an instance of this device definition, use the following command:
    </p>

    <pre>$ virsh nodedev-start &lt;device-name&gt;</pre>
    <p>
      Active mediated device instances can be stopped using <code>virsh
      nodedev-destroy</code>, and persistent device definitions can be removed
      using <code>virsh nodedev-undefine</code>.
    </p>

    <p>
      If a mediated device is defined persistently, it can also be set to be
      automatically started whenever the host reboots or when the parent device
      becomes available. In order to autostart a mediated device, use the
      following command:
    </p>

    <pre>$ virsh nodedev-autostart &lt;device-name&gt;</pre>
  </body>
</html>