libvirt/docs/pci-addresses.rst
Laine Stump 956e1ca6aa tests: remove explicit <driver name='vfio'/> from hostdev test cases
The long-deprecated use of <driver name='vfio|xen|kvm'/> in domain xml
for <hostdev> devices was only ever necessary during the period when
libvirt (and the Linux kernel) supported both VFIO and "legacy KVM"
styles of hostdev device assignment for QEMU. This became pointless
many years ago when legacy KVM device assignment was removed from the
kernel, and support for that style of device assignment was completely
disabled in the libvirt source in 2019 (commit
v5.6.0-316-g2e7225ea8c).

Nevertheless, there were instances of <driver name='vfio'/> in the
unit test data that were then (unnecessarily) propagated to several
more tests over the years. This patch cleans out those unnecessary
explicit settings of driver name='vfio' in all QEMU unit test data,
proving that the attribute is no longer (externally) needed. (A later
patch which adds a 2nd attribute to the <driver> element will include
a test case that explicitly exercises the driver name attribute).

Signed-off-by: Laine Stump <laine@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
2024-01-07 23:59:00 -05:00

9.5 KiB

PCI addresses in domain XML and guest OS

Looking at the configuration for a guest, it would be reasonable to expect that each PCI device would show up in the guest OS with a PCI address that matches the one present in the corresponding <address> element of the domain XML, but that's not guaranteed to happen and will in fact not be the case in all but the simplest scenarios.

Simple cases

When the PCI topology of the VM is very simple, the PCI addresses will usually match.

For example, the domain XML snippet

<controller type='pci' index='0' model='pcie-root'/>
<controller type='pci' index='1' model='pcie-root-port'>
  <model name='pcie-root-port'/>
  <target chassis='1' port='0x8'/>
  <address type='pci' domain='0x0000' bus='0x00' slot='0x01' function='0x0'/>
</controller>
<interface type='network'>
  <source network='default'/>
  <model type='virtio'/>
  <address type='pci' domain='0x0000' bus='0x01' slot='0x00' function='0x0'/>
</interface>

will result in the PCI topology

0000:00:00.0 Host bridge: Intel Corporation 82G33/G31/P35/P31 Express DRAM Controller
0000:00:01.0 PCI bridge: Red Hat, Inc. QEMU PCIe Root port
0000:01:00.0 Ethernet controller: Red Hat, Inc. Virtio network device (rev 01)

showing up in the guest OS.

The PCI address of the virtio-net adapter, 0000:01:00.0, is the same in both cases, so there's no confusion.

More complex cases

In more complex cases, the PCI address visible in the domain XML will correlate to the one seen by the guest OS in a less obvious way.

pcie-expander-bus

This fairly uncommon device, which can be used with x86_64/q35 guests, will help illustrate one such scenario.

For example, the domain XML snippet

<controller type='pci' index='0' model='pcie-root'/>
<controller type='pci' index='1' model='pcie-expander-bus'>
  <model name='pxb-pcie'/>
  <target busNr='254'/>
  <address type='pci' domain='0x0000' bus='0x00' slot='0x01' function='0x0'/>
</controller>
<controller type='pci' index='2' model='pcie-root-port'>
  <model name='pcie-root-port'/>
  <target chassis='2' port='0x0'/>
  <address type='pci' domain='0x0000' bus='0x01' slot='0x00' function='0x0'/>
</controller>
<interface type='network'>
  <source network='default'/>
  <model type='virtio'/>
  <address type='pci' domain='0x0000' bus='0x02' slot='0x00' function='0x0'/>
</interface>

will result in the PCI topology

0000:00:00.0 Host bridge: Intel Corporation 82G33/G31/P35/P31 Express DRAM Controller
0000:00:01.0 Host bridge: Red Hat, Inc. QEMU PCIe Expander bridge
0000:fe:00.0 PCI bridge: Red Hat, Inc. QEMU PCIe Root port
0000:ff:00.0 Ethernet controller: Red Hat, Inc. Virtio network device (rev 01)

showing up in the guest OS.

This time the addresses don't match: this is because the busNr property for the pcie-expander-bus controller causes it to show up as bus 254 (0xfe in hexadecimal) instead of bus 1 as one might expect based on its index property.

How can the domain XML shown above work at all, then? Surely the pcie-root-port controller and the virtio-net adapter should use bus=0xfe and bus=0xff respectively for the configuration to be accepted by libvirt?

As it turns out, that's not the case. The reason for this is that QEMU, and consequently libvirt, uses the bus property of a device's PCI address only to match it with the PCI controller that has the same index property, and not to set the actual PCI address, which is decided by the guest OS.

So, by looking at the XML snippet above, we can see that the virtio-net adapter plugs into the pcie-root-port controller, which plugs into the pcie-expander-bus controller, which plugs into pcie-root: the guest OS sees the same topology, but assigns different PCI addresses to some of its component.

The takeaway is that the relationship between controllers are the very same whether you look at the domain XML or at the guest OS, but the actual PCI addresses are not guaranteed to match and in fact, except for the very simplest cases, they usually will not.

spapr-pci-host-bridge

This device, which is unique to ppc64/pseries guests, will help illustrate another scenario.

For example, the domain XML snippet

<controller type='pci' index='0' model='pci-root'>
   <model name='spapr-pci-host-bridge'/>
   <target index='0'/>
 </controller>
 <controller type='pci' index='1' model='pci-root'>
   <model name='spapr-pci-host-bridge'/>
   <target index='1'/>
 </controller>
 <interface type='network'>
   <source network='default'/>
   <model type='virtio'/>
   <address type='pci' domain='0x0000' bus='0x01' slot='0x01' function='0x0'/>
 </interface>

will result in the PCI topology

0001:00:01.0 Ethernet controller: Red Hat, Inc. Virtio network device

showing up in the guest OS. Note that the two spapr-pci-host-bridge controllers are not listed.

This time, in addition to the bus not matching just like in the previous example, the interesting part is that the domain doesn't match either: this is because each spapr-pci-host-bridge controller creates a separate PCI domain.

Once again, while the PCI addresses seen in the domain XML and those seen by the guest OS do not match, the relationships between the various devices are preserved.

zPCI addresses

For s390x machines, PCI addresses are handled yet differently. No topology information is relayed in the PCI addresses; instead, the fid and uid elements of the zpci device convey information. In the simplest case, the following XML snippet

<controller type='pci' index='0' model='pci-root'/>
<controller type='pci' index='1' model='pci-bridge'>
  <model name='pci-bridge'/>
  <target chassisNr='1'/>
  <address type='pci' domain='0x0000' bus='0x00' slot='0x01' function='0x0'>
    <zpci uid='0x0001' fid='0x00000000'/>
  </address>
</controller>
<interface type='bridge'>
  <source bridge='virbr0'/>
  <model type='virtio'/>
  <address type='pci' domain='0x0000' bus='0x01' slot='0x01' function='0x0'>
    <zpci uid='0x0007' fid='0x00000003'/>
  </address>
</interface>

will result in the following in a Linux guest:

0007:00:00.0 Ethernet controller: Red Hat, Inc. Virtio network device

Note that the PCI bridge is not visible in the guest; s390x always has a flat topology. The PCI address in the guest is generated from the information provided via the zpci element: more specifically, uid is used as the PCI domain. fid doesn't appear in the PCI address itself, but it will be used in sysfs (/sys/bus/pci/slots/$fid/...).

Any changes in the PCI address are not visible in the guest; replacing the PCI address for the virtio-net device with

<address type='pci' domain='0x0000' bus='0x01' slot='0x06' function='0x4'>

will result in the exactly same view in the guest, as the fid and uid values in the zpci element remain unchanged.

Device assignment

When using VFIO to assign host devices to a guest, an additional caveat to keep in mind that the guest OS will base its decisions upon the target address (guest side) rather than the source address (host side).

For example, the domain XML snippet

<hostdev mode='subsystem' type='pci' managed='yes'>
  <source>
    <address domain='0x0001' bus='0x08' slot='0x00' function='0x0'/>
  </source>
  <address type='pci' domain='0x0000' bus='0x00' slot='0x01' function='0x0'/>
</hostdev>

will result in the device showing up as 0000:00:01.0 in the guest OS rather than as 0001:08:00.1, which is the address of the device on the host.

Of course, all the rules and behaviors described above still apply.

Reserved addresses

Due to some historical reasons hypervisors might expect some PCI devices to appear at certain addresses instead of 'random' ones. For QEMU this is machine type and guest architecture dependent. But to give you at least a gist here is list of reserved PCI addresses:

For the x86_64 architecture's I440FX-based machine types the following devices are hard coded into QEMU and can't be moved or eliminated:

0000:00:00.0 Host bridge
0000:00:01.0 ISA bridge
0000:00:01.1 primary IDE controller
0000:00:01.2 PIIX3 USB controller
0000:00:01.3 ACPI (power management) and SMBus controller

The following addresses will be used as default ones for the corresponding devices (if the address is free or a different address wasn't provided for the device). It is okay to use this address for any other device.

0000:00:02.0 primary video card
For the x86_64

architecture's Q35-based machine types the following

devices are ha

rd coded into QEMU and can't be moved or eliminated:

============ =======================
0000:00:00.0 Host bridge
0000:00:1f.2 primary SATA controller
0000:00:1f.0 ISA bridge
0000:00:1f.3 SMBus

The following addresses will be used as default ones for the corresponding devices (if the address is free or a different address wasn't provided for the device) because that's how real Q35 would do it:

0000:00:1a.0 USB2 controller
0000:00:1b.0 ICH9 sound chip
0000:00:1d.0 USB2 controller