CPU model configuration for QEMU/KVM on x86 hosts

Posted: June 29th, 2018 | Author: | Filed under: Fedora, libvirt, OpenStack, Security, Virt Tools | Tags: , , , , , , , | 1 Comment »

With the various CPU hardware vulnerabilities reported this year, guest CPU configuration is now a security critical task. This blog post contains content I’ve written that is on its way to become part of the QEMU documentation.

QEMU / KVM virtualization supports two ways to configure CPU models

Host passthrough
This passes the host CPU model features, model, stepping, exactly to the guest. Note that KVM may filter out some host CPU model features if they cannot be supported with virtualization. Live migration is unsafe when this mode is used as libvirt / QEMU cannot guarantee a stable CPU is exposed to the guest across hosts. This is the recommended CPU to use, provided live migration is not required.
Named model
QEMU comes with a number of predefined named CPU models, that typically refer to specific generations of hardware released by Intel and AMD. These allow the guest VMs to have a degree of isolation from the host CPU, allowing greater flexibility in live migrating between hosts with differing hardware.

In both cases, it is possible to optionally add or remove individual CPU features, to alter what is presented to the guest by default.

Libvirt supports a third way to configure CPU models known as “Host model”. This uses the QEMU “Named model” feature, automatically picking a CPU model that is similar the host CPU, and then adding extra features to approximate the host model as closely as possible. This does not guarantee the CPU family, stepping, etc will precisely match the host CPU, as they would with “Host passthrough”, but gives much of the benefit of passthrough, while making live migration safe.

Recommendations for KVM CPU model configuration on x86 hosts

The information that follows provides recommendations for configuring CPU models on x86 hosts. The goals are to maximise performance, while protecting guest OS against various CPU hardware flaws, and optionally enabling live migration between hosts with hetergeneous CPU models.

Preferred CPU models for Intel x86 hosts

The following CPU models are preferred for use on Intel hosts. Administrators / applications are recommended to use the CPU model that matches the generation of the host CPUs in use. In a deployment with a mixture of host CPU models between machines, if live migration compatibility is required, use the newest CPU model that is compatible across all desired hosts.

Intel Xeon Processor (Skylake, 2016)
Intel Core Processor (Skylake, 2015)
Intel Core Processor (Broadwell, 2014)
Intel Core Processor (Haswell, 2013)
Intel Xeon E3-12xx v2 (Ivy Bridge, 2012)
Intel Xeon E312xx (Sandy Bridge, 2011)
Westmere E56xx/L56xx/X56xx (Nehalem-C, 2010)
Intel Core i7 9xx (Nehalem Class Core i7, 2008)
Intel Core 2 Duo P9xxx (Penryn Class Core 2, 2007)
Intel Celeron_4x0 (Conroe/Merom Class Core 2, 2006)

Important CPU features for Intel x86 hosts

The following are important CPU features that should be used on Intel x86 hosts, when available in the host CPU. Some of them require explicit configuration to enable, as they are not included by default in some, or all, of the named CPU models listed above. In general all of these features are included if using “Host passthrough” or “Host model”.

Recommended to mitigate the cost of the Meltdown (CVE-2017-5754) fix. Included by default in Haswell, Broadwell & Skylake Intel CPU models. Should be explicitly turned on for Westmere, SandyBridge, and IvyBridge Intel CPU models. Note that some desktop/mobile Westmere CPUs cannot support this feature.
Required to enable the Spectre (CVE-2017-5753 and CVE-2017-5715) fix, in cases where retpolines are not sufficient. Included by default in Intel CPU models with -IBRS suffix. Must be explicitly turned on for Intel CPU models without -IBRS suffix. Requires the host CPU microcode to support this feature before it can be used for guest CPUs.
Required to enable the CVE-2018-3639 fix. Not included by default in any Intel CPU model. Must be explicitly turned on for all Intel CPU models. Requires the host CPU microcode to support this feature before it can be used for guest CPUs.
Recommended to allow guest OS to use 1GB size pages.Not included by default in any Intel CPU model. Should be explicitly turned on for all Intel CPU models. Note that not all CPU hardware will support this feature.

Preferred CPU models for AMD x86 hosts

The following CPU models are preferred for use on Intel hosts. Administrators / applications are recommended to use the CPU model that matches the generation of the host CPUs in use. In a deployment with a mixture of host CPU models between machines, if live migration compatibility is required, use the newest CPU model that is compatible across all desired hosts.

AMD EPYC Processor (2017)
AMD Opteron 63xx class CPU (2012)
AMD Opteron 62xx class CPU (2011)
AMD Opteron 23xx (Gen 3 Class Opteron, 2009)
AMD Opteron 22xx (Gen 2 Class Opteron, 2006)
AMD Opteron 240 (Gen 1 Class Opteron, 2004)

Important CPU features for AMD x86 hosts

The following are important CPU features that should be used on AMD x86 hosts, when available in the host CPU. Some of them require explicit configuration to enable, as they are not included by default in some, or all, of the named CPU models listed above. In general all of these features are included if using “Host passthrough” or “Host model”.

Required to enable the Spectre (CVE-2017-5753 and CVE-2017-5715) fix, in cases where retpolines are not sufficient. Included by default in AMD CPU models with -IBPB suffix. Must be explicitly turned on for AMD CPU models without -IBPB suffix. Requires the host CPU microcode to support this feature before it can be used for guest CPUs.
Required to enable the CVE-2018-3639 fix. Not included by default in any AMD CPU model. Must be explicitly turned on for all AMD CPU models. This should be provided to guests, even if amd-ssbd is also provided, for maximum guest compatibility. Note for some QEMU / libvirt versions, this must be force enabled when when using “Host model”, because this is a virtual feature that doesn’t exist in the physical host CPUs.
Required to enable the CVE-2018-3639 fix. Not included by default in any AMD CPU model. Must be explicitly turned on for all AMD CPU models. This provides higher performance than virt-ssbd so should be exposed to guests whenever available in the host. virt-ssbd should none the less also be exposed for maximum guest compatability as some kernels only know about virt-ssbd.
Recommended to indicate the host is not vulnerable CVE-2018-3639. Not included by default in any AMD CPU model. Future hardware genarations of CPU will not be vulnerable to CVE-2018-3639, and thus the guest should be told not to enable its mitigations, by exposing amd-no-ssb. This is mutually exclusive with virt-ssbd and amd-ssbd.
Recommended to allow guest OS to use 1GB size pages. Not included by default in any AMD CPU model. Should be explicitly turned on for all AMD CPU models. Note that not all CPU hardware will support this feature.

Default x86 CPU models

The default QEMU CPU models are designed such that they can run on all hosts. If an application does not wish to do perform any host compatibility checks before launching guests, the default is guaranteed to work.

The default CPU models will, however, leave the guest OS vulnerable to various CPU hardware flaws, so their use is strongly discouraged. Applications should follow the earlier guidance to setup a better CPU configuration, with host passthrough recommended if live migration is not needed.

QEMU Virtual CPU version 2.5+ (32 & 64 bit variants). qemu64 is used for x86_64 guests and qemu32 is used for i686 guests, when no -cpu argument is given to QEMU, or no <cpu> is provided in libvirt XML.

Other non-recommended x86 CPUs

The following CPUs models are compatible with most AMD and Intel x86 hosts, but their usage is discouraged, as they expose a very limited featureset, which prevents guests having optimal performance.

Common KVM processor (32 & 64 bit variants). Legacy models just for historical compatibility with ancient QEMU versions.
Various very old x86 CPU models, mostly predating the introduction of hardware assisted virtualization, that should thus not be required for running virtual machines.

Syntax for configuring CPU models

The example below illustrate the approach to configuring the various CPU models / features in QEMU and libvirt

QEMU command line

Host passthrough
   $ qemu-system-x86_64 -cpu host

With feature customization:

   $ qemu-system-x86_64 -cpu host,-vmx,...
Named CPU models
   $ qemu-system-x86_64 -cpu Westmere

With feature customization:

   $ qemu-system-x86_64 -cpu Westmere,+pcid,...

Libvirt guest XML

Host passthrough
   <cpu mode='host-passthrough'/>

With feature customization:

   <cpu mode='host-passthrough'>
       <feature name="vmx" policy="disable"/>
Host model
   <cpu mode='host-model'/>

With feature customization:

   <cpu mode='host-model'>
       <feature name="vmx" policy="disable"/>
Named model
   <cpu mode='custom'>
       <model name="Westmere"/>

With feature customization:

   <cpu mode='custom'>
       <model name="Westmere"/>
       <feature name="pcid" policy="require"/>


Full coverage of libvirt XML schemas achieved in libvirt-go-xml

Posted: December 7th, 2017 | Author: | Filed under: Coding Tips, Fedora, libvirt, OpenStack, Virt Tools | Tags: , | No Comments »

In recent times I have been aggressively working to expand the coverage of libvirt XML schemas in the libvirt-go-xml project. Today this work has finally come to a conclusion, when I achieved what I believe to be effectively 100% coverage of all of the libvirt XML schemas. More on this later, but first some background on Go and XML….

For those who aren’t familiar with Go, the core library’s encoding/xml module provides a very easy way to consume and produce XML documents in Go code. You simply define a set of struct types and annotate their fields to indicate what elements & attributes each should map to. For example, given the Go structs:

type Person struct {
    XMLName xml.Name `xml:"person"`
    Name string `xml:"name,attr"`
    Age string `xml:"age,attr"` 
    Home *Address `xml:"home"`
    Office *Address `xml:"office"`
type Address struct { 
    Street string `xml:"street"`
    City string `xml:"city"` 

You can parse/format XML documents looking like

<person name="Joe Blogs" age="24">
    <street>Some where</street><city>London</city>
    <street>Some where else</street><city>London</city>

Other programming languages I’ve used required a great deal more work when dealing with XML. For parsing, there’s typically a choice between an XML stream based parser where you have to react to tokens as they’re parsed and stuff them into structs, or a DOM object hierarchy from which you then have to pull data out into your structs. For outputting XML, apps either build up a DOM object hierarchy again, or dynamically format the XML document incrementally. Whichever approach is taken, it generally involves writing alot of tedious & error prone boilerplate code. In most cases, the Go encoding/xml module eliminates all the boilerplate code, only requiring the data type defintions. This really makes dealing with XML a much more enjoyable experience, because you effectively don’t deal with XML at all! There are some exceptions to this though, as the simple annotations can’t capture every nuance of many XML documents. For example, integer values are always parsed & formatted in base 10, so extra work is needed for base 16. There’s also no concept of unions in Go, or the XML annotations. In these edge cases custom marshaling / unmarshalling methods need to be written. BTW, this approach to XML is also taken for other serialization formats including JSON and YAML too, with one struct field able to have many annotations so it can be serialized to a range of formats.

Back to the point of the blog post, when I first started writing Go code using libvirt it was immediately obvious that everyone using libvirt from Go would end up re-inventing the wheel for XML handling. Thus about 1 year ago, I created the libvirt-go-xml project whose goal is to define a set of structs that can handle documents in every libvirt public XML schema. Initially the level of coverage was fairly light, and over the past year 18 different contributors have sent patches to expand the XML coverage in areas that their respective applications touched. It was clear, however, that taking an incremental approach would mean that libvirt-go-xml is forever trailing what libvirt itself supports. It needed an aggressive push to achieve 100% coverage of the XML schemas, or as near as practically identifiable.

Alongside each set of structs we had also been writing unit tests with a set of structs populated with data, and a corresponding expected XML document. The idea for writing the tests was that the author would copy a snippet of XML from a known good source, and then populate the structs that would generate this XML. In retrospect this was not a scalable approach, because there is an enourmous range of XML documents that libvirt supports. A further complexity is that Go doesn’t generate XML documents in the exact same manner. For example, it never generates self-closing tags, instead always outputting a full opening & closing pair. This is semantically equivalent, but makes a plain string comparison of two XML documents impractical in the general case.

Considering the need to expand the XML coverage, and provide a more scalable testing approach, I decided to change approach. The libvirt.git tests/ directory currently contains 2739 XML documents that are used to validate libvirt’s own native XML parsing & formatting code. There is no better data set to use for validating the libvirt-go-xml coverage than this. Thus I decided to apply a round-trip testing methodology. The libvirt-go-xml code would be used to parse the sample XML document from libvirt.git, and then immediately serialize them back into a new XML document. Both the original and new XML documents would then be parsed generically to form a DOM hierarchy which can be compared for equivalence. Any place where documents differ would cause the test to fail and print details of where the problem is. For example:

$ go test -tags xmlroundtrip
--- FAIL: TestRoundTrip (1.01s)
	xml_test.go:384: testdata/libvirt/tests/vircaps2xmldata/vircaps-aarch64-basic.xml: \
            /capabilities[0]/host[0]/topology[0]/cells[0]/cell[0]/pages[0]: \
            element in expected XML missing in actual XML

This shows the filename that failed to correctly roundtrip, and the position within the XML tree that didn’t match. Here the NUMA cell topology has a ‘<pages>‘  element expected but not present in the newly generated XML. Now it was simply a matter of running the roundtrip test over & over & over & over & over & over & over……….& over & over & over, adding structs / fields for each omission that the test identified.

After doing this for some time, libvirt-go-xml now has 586 structs defined containing 1816 fields, and has certified 100% coverage of all libvirt public XML schemas. Of course when I say 100% coverage, this is probably a lie, as I’m blindly assuming that the libvirt.git test suite has 100% coverage of all its own XML schemas. This is certainly a goal, but I’m confident there are cases where libvirt itself is missing test coverage. So if any omissions are identified in libvirt-go-xml, these are likely omissions in libvirt’s own testing.

On top of this, the XML roundtrip test is set to run in the libvirt jenkins and travis CI systems, so as libvirt extends its XML schemas, we’ll get build failures in libvirt-go-xml and thus know to add support there to keep up.

In expanding the coverage of XML schemas, a number of non-trivial changes were made to existing structs  defined by libvirt-go-xml. These were mostly in places where we have to handle a union concept defined by libvirt. Typically with libvirt an element will have a “type” attribute, whose value then determines what child elements are permitted. Previously we had been defining a single struct, whose fields represented all possible children across all the permitted type values. This did not scale well and gave the developer no clue what content is valid for each type value. In the new approach, for each distinct type attribute value, we now define a distinct Go struct to hold the contents. This will cause API breakage for apps already using libvirt-go-xml, but on balance it is worth it get a better structure over the long term. There were also cases where a child XML element previously represented a single value and this was mapped to a scalar struct field. Libvirt then added one or more attributes on this element, meaning the scalar struct field had to turn into a struct field that points to another struct. These kind of changes are unavoidable in any nice manner, so while we endeavour not to gratuitously change currently structs, if the libvirt XML schema gains new content, it might trigger further changes in the libvirt-go-xml structs that are not 100% backwards compatible.

Since we are now tracking libvirt.git XML schemas, going forward we’ll probably add tags in the libvirt-go-xml repo that correspond to each libvirt release. So for app developers we’ll encourage use of Go vendoring to pull in a precise version of libvirt-go-xml instead of blindly tracking master all the time.

Full colour emojis in virtual machine names in Fedora 27

Posted: December 1st, 2017 | Author: | Filed under: Coding Tips, Fedora, libvirt, OpenStack, Virt Tools | Tags: , , , , | 1 Comment »

Quite by chance today I discovered that Fedora 27 can display full colour glyphs for unicode characters that correspond to emojis, when the terminal displaying my mutt mail reader displayed someone’s name with a full colour glyph showing stars:

Mutt in GNOME terminal rendering color emojis in sender name

Chatting with David Gilbert on IRC I learnt that this is a new feature in Fedora 27 GNOME, thanks to recent work in the GTK/Pango stack. David then pointed out this works in libvirt, so I thought I would illustrate it.

Virtual machine name with full colour emojis rendered

No special hacks were required to do this, I simply entered the emojis as the virtual machine name when creating it from virt-manager’s wizard

Virtual machine name with full colour emojis rendered

As mentioned previously, GNOME terminal displays colour emojis, so these virtual machine names appear nicely when using virsh and other command line tools

Virtual machine name rendered with full colour emojis in terminal commands

The more observant readers will notice that the command line args have a bug as the snowman in the machine name is incorrectly rendered in the process listing. The actual data in /proc/$PID/cmdline is correct, so something about the “ps” command appears to be mangling it prior to output. It isn’t simply a font problem because other comamnds besides “ps” render properly, and if you grep the “ps” output for the snowman emoji no results are displayed.

ANNOUNCE: libosinfo 1.1.0 release

Posted: August 15th, 2017 | Author: | Filed under: Fedora, libvirt, OpenStack, Virt Tools | Tags: , | No Comments »

I am happy to announce a new release of libosinfo version 1.1.0 is now available, signed with key DAF3 A6FD B26B 6291 2D0E 8E3F BE86 EBB4 1510 4FDF (4096R). All historical releases are available from the project download page.

Changes in this release include:

  • Force UTF-8 locale for new glib-mkenums
  • Avoid python warnings in example program
  • Misc test suite updates
  • Fix typo in error messages
  • Remove ISO header string padding
  • Disable bogus gcc warning about unsafe loop optimizations
  • Remove reference to fedorahosted.org
  • Don’t hardcode /usr/bin/perl, use /usr/bin/env
  • Support eject-after-install parameter in OsinfoMedia
  • Fix misc warnings in docs
  • Fix error propagation when loading DB
  • Add usb.ids / pci.ids locations for FreeBSD
  • Don’t include private headers in gir/vapi generation

Thanks to everyone who contributed towards this release.

Setting up a nested KVM guest for developing & testing PCI device assignment with NUMA

Posted: February 16th, 2017 | Author: | Filed under: Coding Tips, Fedora, libvirt, OpenStack, Virt Tools | Tags: , , , , , | No Comments »

Over the past few years OpenStack Nova project has gained support for managing VM usage of NUMA, huge pages and PCI device assignment. One of the more challenging aspects of this is availability of hardware to develop and test against. In the ideal world it would be possible to emulate everything we need using KVM, enabling developers / test infrastructure to exercise the code without needing access to bare metal hardware supporting these features. KVM has long has support for emulating NUMA topology in guests, and guest OS can use huge pages inside the guest. What was missing were pieces around PCI device assignment, namely IOMMU support and the ability to associate NUMA nodes with PCI devices. Co-incidentally a QEMU community member was already working on providing emulation of the Intel IOMMU. I made a request to the Red Hat KVM team to fill in the other missing gap related to NUMA / PCI device association. To do this required writing code to emulate a PCI/PCI-E Expander Bridge (PXB) device, which provides a light weight host bridge that can be associated with a NUMA node. Individual PCI devices are then attached to this PXB instead of the main PCI host bridge, thus gaining affinity with a NUMA node. With this, it is now possible to configure a KVM guest such that it can be used as a virtual host to test NUMA, huge page and PCI device assignment integration. The only real outstanding gap is support for emulating some kind of SRIOV network device, but even without this, it is still possible to test most of the Nova PCI device assignment logic – we’re merely restricted to using physical functions, no virtual functions. This blog posts will describe how to configure such a virtual host.

First of all, this requires very new libvirt & QEMU to work, specifically you’ll want libvirt >= 2.3.0 and QEMU 2.7.0. We could technically support earlier QEMU versions too, but that’s pending on a patch to libvirt to deal with some command line syntax differences in QEMU for older versions. No currently released Fedora has new enough packages available, so even on Fedora 25, you must enable the “Virtualization Preview” repository on the physical host to try this out – F25 has new enough QEMU, so you just need a libvirt update.

# curl --output /etc/yum.repos.d/fedora-virt-preview.repo https://fedorapeople.org/groups/virt/virt-preview/fedora-virt-preview.repo
# dnf upgrade

For sake of illustration I’m using Fedora 25 as the OS inside the virtual guest, but any other Linux OS will do just fine. The initial task is to install guest with 8 GB of RAM & 8 CPUs using virt-install

# cd /var/lib/libvirt/images
# wget -O f25x86_64-boot.iso https://download.fedoraproject.org/pub/fedora/linux/releases/25/Server/x86_64/os/images/boot.iso
# virt-install --name f25x86_64  \
    --file /var/lib/libvirt/images/f25x86_64.img --file-size 20 \
    --cdrom f25x86_64-boot.iso --os-type fedora23 \
    --ram 8000 --vcpus 8 \

The guest needs to use host CPU passthrough to ensure the guest gets to see VMX, as well as other modern instructions and have 3 virtual NUMA nodes. The first guest NUMA node will have 4 CPUs and 4 GB of RAM, while the second and third NUMA nodes will each have 2 CPUs and 2 GB of RAM. We are just going to let the guest float freely across host NUMA nodes since we don’t care about performance for dev/test, but in production you would certainly pin each guest NUMA node to a distinct host NUMA node.

    --cpu host,cell0.id=0,cell0.cpus=0-3,cell0.memory=4096000,\
               cell2.id=2,cell2.cpus=6-7,cell2.memory=2048000 \

QEMU emulates various different chipsets and historically for x86, the default has been to emulate the ancient PIIX4 (it is 20+ years old dating from circa 1995). Unfortunately this is too ancient to be able to use the Intel IOMMU emulation with, so it is neccessary to tell QEMU to emulate the marginally less ancient chipset Q35 (it is only 9 years old, dating from 2007).

    --machine q35

The complete virt-install command line thus looks like

# virt-install --name f25x86_64  \
    --file /var/lib/libvirt/images/f25x86_64.img --file-size 20 \
    --cdrom f25x86_64-boot.iso --os-type fedora23 \
    --ram 8000 --vcpus 8 \
    --cpu host,cell0.id=0,cell0.cpus=0-3,cell0.memory=4096000,\
               cell2.id=2,cell2.cpus=6-7,cell2.memory=2048000 \
    --machine q35

Once the installation is completed, shut down this guest since it will be necessary to make a number of changes to the guest XML configuration to enable features that virt-install does not know about, using “virsh edit“. With the use of Q35, the guest XML should initially show three PCI controllers present, a “pcie-root”, a “dmi-to-pci-bridge” and a “pci-bridge”

<controller type='pci' index='0' model='pcie-root'/>
<controller type='pci' index='1' model='dmi-to-pci-bridge'>
  <model name='i82801b11-bridge'/>
  <address type='pci' domain='0x0000' bus='0x00' slot='0x1e' function='0x0'/>
<controller type='pci' index='2' model='pci-bridge'>
  <model name='pci-bridge'/>
  <target chassisNr='2'/>
  <address type='pci' domain='0x0000' bus='0x01' slot='0x00' function='0x0'/>

PCI endpoint devices are not themselves associated with NUMA nodes, rather the bus they are connected to has affinity. The default pcie-root is not associated with any NUMA node, but extra PCI-E Expander Bridge controllers can be added and associated with a NUMA node. So while in edit mode, add the following to the XML config

<controller type='pci' index='3' model='pcie-expander-bus'>
  <target busNr='180'>
  <address type='pci' domain='0x0000' bus='0x00' slot='0x02' function='0x0'/>
<controller type='pci' index='4' model='pcie-expander-bus'>
  <target busNr='200'>
  <address type='pci' domain='0x0000' bus='0x00' slot='0x03' function='0x0'/>
<controller type='pci' index='5' model='pcie-expander-bus'>
  <target busNr='220'>
  <address type='pci' domain='0x0000' bus='0x00' slot='0x04' function='0x0'/>

It is not possible to plug PCI endpoint devices directly into the PXB, so the next step is to add PCI-E root ports into each PXB – we’ll need one port per device to be added, so 9 ports in total. This is where the requirement for libvirt >= 2.3.0 – earlier versions mistakenly prevented you adding more than one root port to the PXB

<controller type='pci' index='6' model='pcie-root-port'>
  <model name='ioh3420'/>
  <target chassis='6' port='0x0'/>
  <alias name='pci.6'/>
  <address type='pci' domain='0x0000' bus='0x03' slot='0x00' function='0x0'/>
<controller type='pci' index='7' model='pcie-root-port'>
  <model name='ioh3420'/>
  <target chassis='7' port='0x8'/>
  <alias name='pci.7'/>
  <address type='pci' domain='0x0000' bus='0x03' slot='0x01' function='0x0'/>
<controller type='pci' index='8' model='pcie-root-port'>
  <model name='ioh3420'/>
  <target chassis='8' port='0x10'/>
  <alias name='pci.8'/>
  <address type='pci' domain='0x0000' bus='0x03' slot='0x02' function='0x0'/>
<controller type='pci' index='9' model='pcie-root-port'>
  <model name='ioh3420'/>
  <target chassis='9' port='0x0'/>
  <alias name='pci.9'/>
  <address type='pci' domain='0x0000' bus='0x04' slot='0x00' function='0x0'/>
<controller type='pci' index='10' model='pcie-root-port'>
  <model name='ioh3420'/>
  <target chassis='10' port='0x8'/>
  <alias name='pci.10'/>
  <address type='pci' domain='0x0000' bus='0x04' slot='0x01' function='0x0'/>
<controller type='pci' index='11' model='pcie-root-port'>
  <model name='ioh3420'/>
  <target chassis='11' port='0x10'/>
  <alias name='pci.11'/>
  <address type='pci' domain='0x0000' bus='0x04' slot='0x02' function='0x0'/>
<controller type='pci' index='12' model='pcie-root-port'>
  <model name='ioh3420'/>
  <target chassis='12' port='0x0'/>
  <alias name='pci.12'/>
  <address type='pci' domain='0x0000' bus='0x05' slot='0x00' function='0x0'/>
<controller type='pci' index='13' model='pcie-root-port'>
  <model name='ioh3420'/>
  <target chassis='13' port='0x8'/>
  <alias name='pci.13'/>
  <address type='pci' domain='0x0000' bus='0x05' slot='0x01' function='0x0'/>
<controller type='pci' index='14' model='pcie-root-port'>
  <model name='ioh3420'/>
  <target chassis='14' port='0x10'/>
  <alias name='pci.14'/>
  <address type='pci' domain='0x0000' bus='0x05' slot='0x02' function='0x0'/>|

Notice that the values in ‘bus‘ attribute on the <address> element is matching the value of the ‘index‘ attribute on the <controller> element of the parent device in the topology. The PCI controller topology now looks like this

pcie-root (index == 0)
  +- dmi-to-pci-bridge (index == 1)
  |    |
  |    +- pci-bridge (index == 2)
  +- pcie-expander-bus (index == 3, numa node == 0)
  |    |
  |    +- pcie-root-port (index == 6)
  |    +- pcie-root-port (index == 7)
  |    +- pcie-root-port (index == 8)
  +- pcie-expander-bus (index == 4, numa node == 1)
  |    |
  |    +- pcie-root-port (index == 9)
  |    +- pcie-root-port (index == 10)
  |    +- pcie-root-port (index == 11)
  +- pcie-expander-bus (index == 5, numa node == 2)
       +- pcie-root-port (index == 12)
       +- pcie-root-port (index == 13)
       +- pcie-root-port (index == 14)

All the existing devices are attached to the “pci-bridge” (the controller with index == 2). The devices we intend to use for PCI device assignment inside the virtual host will be attached to the new “pcie-root-port” controllers. We will provide 3 e1000 per NUMA node, so that’s 9 devices in total to add

<interface type='user'>
  <mac address='52:54:00:7e:6e:c6'/>
  <model type='e1000e'/>
  <address type='pci' domain='0x0000' bus='0x06' slot='0x00' function='0x0'/>
<interface type='user'>
  <mac address='52:54:00:7e:6e:c7'/>
  <model type='e1000e'/>
  <address type='pci' domain='0x0000' bus='0x07' slot='0x00' function='0x0'/>
<interface type='user'>
  <mac address='52:54:00:7e:6e:c8'/>
  <model type='e1000e'/>
  <address type='pci' domain='0x0000' bus='0x08' slot='0x00' function='0x0'/>
<interface type='user'>
  <mac address='52:54:00:7e:6e:d6'/>
  <model type='e1000e'/>
  <address type='pci' domain='0x0000' bus='0x09' slot='0x00' function='0x0'/>
<interface type='user'>
  <mac address='52:54:00:7e:6e:d7'/>
  <model type='e1000e'/>
  <address type='pci' domain='0x0000' bus='0x0a' slot='0x00' function='0x0'/>
<interface type='user'>
  <mac address='52:54:00:7e:6e:d8'/>
  <model type='e1000e'/>
  <address type='pci' domain='0x0000' bus='0x0b' slot='0x00' function='0x0'/>
<interface type='user'>
  <mac address='52:54:00:7e:6e:e6'/>
  <model type='e1000e'/>
  <address type='pci' domain='0x0000' bus='0x0c' slot='0x00' function='0x0'/>
<interface type='user'>
  <mac address='52:54:00:7e:6e:e7'/>
  <model type='e1000e'/>
  <address type='pci' domain='0x0000' bus='0x0d' slot='0x00' function='0x0'/>
<interface type='user'>
  <mac address='52:54:00:7e:6e:e8'/>
  <model type='e1000e'/>
  <address type='pci' domain='0x0000' bus='0x0e' slot='0x00' function='0x0'/>

Note that we’re using the “user” networking, aka SLIRP. Normally one would never want to use SLIRP but we don’t care about actually sending traffic over these NICs, and so using SLIRP avoids polluting our real host with countless TAP devices.

The final configuration change is to simply add the Intel IOMMU device

<iommu model='intel'/>

It is a capability integrated into the chipset, so it does not need any <address> element of its own. At this point, save the config and start the guest once more. Use the “virsh domifaddrs” command to discover the IP address of the guest’s primary NIC and ssh into it.

# virsh domifaddr f25x86_64
 Name       MAC address          Protocol     Address
 vnet0      52:54:00:10:26:7e    ipv4

# ssh root@

We can now do some sanity check that everything visible in the guest matches what was enabled in the libvirt XML config in the host. For example, confirm the NUMA topology shows 3 nodes

# dnf install numactl
# numactl --hardware
available: 3 nodes (0-2)
node 0 cpus: 0 1 2 3
node 0 size: 3856 MB
node 0 free: 3730 MB
node 1 cpus: 4 5
node 1 size: 1969 MB
node 1 free: 1813 MB
node 2 cpus: 6 7
node 2 size: 1967 MB
node 2 free: 1832 MB
node distances:
node   0   1   2 
  0:  10  20  20 
  1:  20  10  20 
  2:  20  20  10 

Confirm that the PCI topology shows the three PCI-E Expander Bridge devices, each with three NICs attached

# lspci -t -v
-+-[0000:dc]-+-00.0-[dd]----00.0  Intel Corporation 82574L Gigabit Network Connection
 |           +-01.0-[de]----00.0  Intel Corporation 82574L Gigabit Network Connection
 |           \-02.0-[df]----00.0  Intel Corporation 82574L Gigabit Network Connection
 +-[0000:c8]-+-00.0-[c9]----00.0  Intel Corporation 82574L Gigabit Network Connection
 |           +-01.0-[ca]----00.0  Intel Corporation 82574L Gigabit Network Connection
 |           \-02.0-[cb]----00.0  Intel Corporation 82574L Gigabit Network Connection
 +-[0000:b4]-+-00.0-[b5]----00.0  Intel Corporation 82574L Gigabit Network Connection
 |           +-01.0-[b6]----00.0  Intel Corporation 82574L Gigabit Network Connection
 |           \-02.0-[b7]----00.0  Intel Corporation 82574L Gigabit Network Connection
 \-[0000:00]-+-00.0  Intel Corporation 82G33/G31/P35/P31 Express DRAM Controller
             +-01.0  Red Hat, Inc. QXL paravirtual graphic card
             +-02.0  Red Hat, Inc. Device 000b
             +-03.0  Red Hat, Inc. Device 000b
             +-04.0  Red Hat, Inc. Device 000b
             +-1d.0  Intel Corporation 82801I (ICH9 Family) USB UHCI Controller #1
             +-1d.1  Intel Corporation 82801I (ICH9 Family) USB UHCI Controller #2
             +-1d.2  Intel Corporation 82801I (ICH9 Family) USB UHCI Controller #3
             +-1d.7  Intel Corporation 82801I (ICH9 Family) USB2 EHCI Controller #1
             +-1e.0-[01-02]----01.0-[02]--+-01.0  Red Hat, Inc Virtio network device
             |                            +-02.0  Intel Corporation 82801FB/FBM/FR/FW/FRW (ICH6 Family) High Definition Audio Controller
             |                            +-03.0  Red Hat, Inc Virtio console
             |                            +-04.0  Red Hat, Inc Virtio block device
             |                            \-05.0  Red Hat, Inc Virtio memory balloon
             +-1f.0  Intel Corporation 82801IB (ICH9) LPC Interface Controller
             +-1f.2  Intel Corporation 82801IR/IO/IH (ICH9R/DO/DH) 6 port SATA Controller [AHCI mode]
             \-1f.3  Intel Corporation 82801I (ICH9 Family) SMBus Controller

The IOMMU support will not be enabled yet as the kernel defaults to leaving it off. To enable it, we must update the kernel command line parameters with grub.

# vi /etc/default/grub
....add "intel_iommu=on"...
# grub2-mkconfig > /etc/grub2.cfg

While intel-iommu device in QEMU can do interrupt remapping, there is no way enable that feature via libvirt at this time. So we need to set a hack for vfio

echo "options vfio_iommu_type1 allow_unsafe_interrupts=1" > \

This is also a good time to install libvirt and KVM inside the guest

# dnf groupinstall "Virtualization"
# dnf install libvirt-client
# rm -f /etc/libvirt/qemu/networks/autostart/default.xml

Note we’re disabling the default libvirt network, since it’ll clash with the IP address range used by this guest. An alternative would be to edit the default.xml to change the IP subnet.

Now reboot the guest. When it comes back up, there should be a /dev/kvm device present in the guest.

# ls -al /dev/kvm
crw-rw-rw-. 1 root kvm 10, 232 Oct  4 12:14 /dev/kvm

If this is not the case, make sure the physical host has nested virtualization enabled for the “kvm-intel” or “kvm-amd” kernel modules.

The IOMMU should have been detected and activated

# dmesg  | grep -i DMAR
[    0.000000] ACPI: DMAR 0x000000007FFE2541 000048 (v01 BOCHS  BXPCDMAR 00000001 BXPC 00000001)
[    0.000000] DMAR: IOMMU enabled
[    0.203737] DMAR: Host address width 39
[    0.203739] DMAR: DRHD base: 0x000000fed90000 flags: 0x1
[    0.203776] DMAR: dmar0: reg_base_addr fed90000 ver 1:0 cap 12008c22260206 ecap f02
[    2.910862] DMAR: No RMRR found
[    2.910863] DMAR: No ATSR found
[    2.914870] DMAR: dmar0: Using Queued invalidation
[    2.914924] DMAR: Setting RMRR:
[    2.914926] DMAR: Prepare 0-16MiB unity mapping for LPC
[    2.915039] DMAR: Setting identity map for device 0000:00:1f.0 [0x0 - 0xffffff]
[    2.915140] DMAR: Intel(R) Virtualization Technology for Directed I/O

The key message confirming everything is good is the last line there – if that’s missing something went wrong – don’t be mislead by the earlier “DMAR: IOMMU enabled” line which merely says the kernel saw the “intel_iommu=on” command line option.

The IOMMU should also have registered the PCI devices into various groups

# dmesg  | grep -i iommu  |grep device
[    2.915212] iommu: Adding device 0000:00:00.0 to group 0
[    2.915226] iommu: Adding device 0000:00:01.0 to group 1
[    5.588723] iommu: Adding device 0000:b5:00.0 to group 14
[    5.588737] iommu: Adding device 0000:b6:00.0 to group 15
[    5.588751] iommu: Adding device 0000:b7:00.0 to group 16

Libvirt meanwhile should have detected all the PCI controllers/devices

# virsh nodedev-list --tree
  +- net_lo_00_00_00_00_00_00
  +- pci_0000_00_00_0
  +- pci_0000_00_01_0
  +- pci_0000_00_02_0
  +- pci_0000_00_03_0
  +- pci_0000_00_04_0
  +- pci_0000_00_1d_0
  |   |
  |   +- usb_usb2
  |       |
  |       +- usb_2_0_1_0
  +- pci_0000_00_1d_1
  |   |
  |   +- usb_usb3
  |       |
  |       +- usb_3_0_1_0
  +- pci_0000_00_1d_2
  |   |
  |   +- usb_usb4
  |       |
  |       +- usb_4_0_1_0
  +- pci_0000_00_1d_7
  |   |
  |   +- usb_usb1
  |       |
  |       +- usb_1_0_1_0
  |       +- usb_1_1
  |           |
  |           +- usb_1_1_1_0
  +- pci_0000_00_1e_0
  |   |
  |   +- pci_0000_01_01_0
  |       |
  |       +- pci_0000_02_01_0
  |       |   |
  |       |   +- net_enp2s1_52_54_00_10_26_7e
  |       |     
  |       +- pci_0000_02_02_0
  |       +- pci_0000_02_03_0
  |       +- pci_0000_02_04_0
  |       +- pci_0000_02_05_0
  +- pci_0000_00_1f_0
  +- pci_0000_00_1f_2
  |   |
  |   +- scsi_host0
  |   +- scsi_host1
  |   +- scsi_host2
  |   +- scsi_host3
  |   +- scsi_host4
  |   +- scsi_host5
  +- pci_0000_00_1f_3
  +- pci_0000_b4_00_0
  |   |
  |   +- pci_0000_b5_00_0
  |       |
  |       +- net_enp181s0_52_54_00_7e_6e_c6
  +- pci_0000_b4_01_0
  |   |
  |   +- pci_0000_b6_00_0
  |       |
  |       +- net_enp182s0_52_54_00_7e_6e_c7
  +- pci_0000_b4_02_0
  |   |
  |   +- pci_0000_b7_00_0
  |       |
  |       +- net_enp183s0_52_54_00_7e_6e_c8
  +- pci_0000_c8_00_0
  |   |
  |   +- pci_0000_c9_00_0
  |       |
  |       +- net_enp201s0_52_54_00_7e_6e_d6
  +- pci_0000_c8_01_0
  |   |
  |   +- pci_0000_ca_00_0
  |       |
  |       +- net_enp202s0_52_54_00_7e_6e_d7
  +- pci_0000_c8_02_0
  |   |
  |   +- pci_0000_cb_00_0
  |       |
  |       +- net_enp203s0_52_54_00_7e_6e_d8
  +- pci_0000_dc_00_0
  |   |
  |   +- pci_0000_dd_00_0
  |       |
  |       +- net_enp221s0_52_54_00_7e_6e_e6
  +- pci_0000_dc_01_0
  |   |
  |   +- pci_0000_de_00_0
  |       |
  |       +- net_enp222s0_52_54_00_7e_6e_e7
  +- pci_0000_dc_02_0
      +- pci_0000_df_00_0
          +- net_enp223s0_52_54_00_7e_6e_e8

And if you look at at specific PCI device, it should report the NUMA node it is associated with and the IOMMU group it is part of

# virsh nodedev-dumpxml pci_0000_df_00_0
  <capability type='pci'>
    <product id='0x10d3'>82574L Gigabit Network Connection</product>
    <vendor id='0x8086'>Intel Corporation</vendor>
    <iommuGroup number='10'>
      <address domain='0x0000' bus='0xdc' slot='0x02' function='0x0'/>
      <address domain='0x0000' bus='0xdf' slot='0x00' function='0x0'/>
    <numa node='2'/>
      <link validity='cap' port='0' speed='2.5' width='1'/>
      <link validity='sta' speed='2.5' width='1'/>

Finally, libvirt should also be reporting the NUMA topology

# virsh capabilities
  <cells num='3'>
    <cell id='0'>
      <memory unit='KiB'>4014464</memory>
      <pages unit='KiB' size='4'>1003616</pages>
      <pages unit='KiB' size='2048'>0</pages>
      <pages unit='KiB' size='1048576'>0</pages>
        <sibling id='0' value='10'/>
        <sibling id='1' value='20'/>
        <sibling id='2' value='20'/>
      <cpus num='4'>
        <cpu id='0' socket_id='0' core_id='0' siblings='0'/>
        <cpu id='1' socket_id='1' core_id='0' siblings='1'/>
        <cpu id='2' socket_id='2' core_id='0' siblings='2'/>
        <cpu id='3' socket_id='3' core_id='0' siblings='3'/>
    <cell id='1'>
      <memory unit='KiB'>2016808</memory>
      <pages unit='KiB' size='4'>504202</pages>
      <pages unit='KiB' size='2048'>0</pages>
      <pages unit='KiB' size='1048576'>0</pages>
        <sibling id='0' value='20'/>
        <sibling id='1' value='10'/>
        <sibling id='2' value='20'/>
      <cpus num='2'>
        <cpu id='4' socket_id='4' core_id='0' siblings='4'/>
        <cpu id='5' socket_id='5' core_id='0' siblings='5'/>
    <cell id='2'>
      <memory unit='KiB'>2014644</memory>
      <pages unit='KiB' size='4'>503661</pages>
      <pages unit='KiB' size='2048'>0</pages>
      <pages unit='KiB' size='1048576'>0</pages>
        <sibling id='0' value='20'/>
        <sibling id='1' value='20'/>
        <sibling id='2' value='10'/>
      <cpus num='2'>
        <cpu id='6' socket_id='6' core_id='0' siblings='6'/>
        <cpu id='7' socket_id='7' core_id='0' siblings='7'/>

Everything should be ready and working at this point, so lets try and install a nested guest, and assign it one of the e1000e PCI devices. For simplicity we’ll just do the exact same install for the nested guest, as we used for the top level guest we’re currently running in. The only difference is that we’ll assign it a PCI device

# cd /var/lib/libvirt/images
# wget -O f25x86_64-boot.iso https://download.fedoraproject.org/pub/fedora/linux/releases/25/Server/x86_64/os/images/boot.iso
# virt-install --name f25x86_64 --ram 2000 --vcpus 8 \
    --file /var/lib/libvirt/images/f25x86_64.img --file-size 10 \
    --cdrom f25x86_64-boot.iso --os-type fedora23 \
    --hostdev pci_0000_df_00_0 --network none

If everything went well, you should now have a nested guest with an assigned PCI device attached to it.

This turned out to be a rather long blog posting, but this is not surprising as we’re experimenting with some cutting edge KVM features trying to emulate quite a complicated hardware setup, that deviates from normal KVM guest setup quite a way. Perhaps in the future virt-install will be able to simplify some of this, but at least for the short-medium term there’ll be a fair bit of work required. The positive thing though is that this has clearly demonstrated that KVM is now advanced enough that you can now reasonably expect to do development and testing of features like NUMA and PCI device assignment inside nested guests.

The next step is to convince someone to add QEMU emulation of an Intel SRIOV network device….volunteers please :-)