Editing Using JACK and PipeWire

__FORCETOC__
QEMU's JACK backend provides the best audio quality out of all the available options (Scream, PulseAudio, etc.). One of the easiest ways to leverage the JACK support is by using PipeWire, which emulates JACK.

First, you must make sure your distro's QEMU is built with JACK support:
* On Debian, this is not the case until QEMU 6.1. This is currently only in unstable and bullseye-backports. You should enable backports if you are using bullseye and upgrade if you are using something older.

Normally, you can run applications with PipeWire's JACK support by running it with <code>pw-jack</code>, but this is not possible when using libvirt. Instead, you need PipeWire to provide drop-in JACK support. To do this:
* on Debian: run <code>sudo cp /usr/share/doc/pipewire/examples/ld.so.conf.d/pipewire-jack-*.conf /etc/ld.so.conf.d/; sudo ldconfig</code>
* on Arch: install the <code>qemu-audio-jack</code> package, which contains the jack driver for qemu, and the <code>pipewire-jack</code> package.

Once this is done, you have two options. You could run your virtual machines under your user, or you could set it up so that the QEMU user has access to your PipeWire socket. The former is easier, while the latter is more secure.

== Run VMs under your user ==

We will assuming your username is <code>example</code>, your user ID is 1000, and your virtual machine is called <code>win10</code>.

You can configure libvirt to run QEMU virtual machines under your user by adding the following line to <code>/etc/libvirt/qemu.conf</code>:

 user = "example"

Then, you can run <code>virsh edit win10</code>, and perform the following changes:

# If the <code><domain></code> line doesn't already contain <code>xmlns:qemu</code>, add <code><nowiki>xmlns:qemu='http://libvirt.org/schemas/domain/qemu/1.0'</nowiki></code>. The line should look like:
#: <syntaxhighlight lang="xml"><domain type='kvm' xmlns:qemu='http://libvirt.org/schemas/domain/qemu/1.0'></syntaxhighlight>
# Add the following block to the <code><devices></code> section:
#: <syntaxhighlight lang="xml">
  <audio id="1" type="jack">
    <input clientName="win10" connectPorts="input-name"/>
    <output clientName="win10" connectPorts="output-name"/>
  </audio>
  <sound model="ich9">
    <alias name="sound0"/>
  </sound>
</syntaxhighlight>
#: <code>input-name</code> and <code>output-name</code> should be replaced with a regular expression that matches the name of the JACK input and output ports, for example <code>connectPorts='Built-in Audio Analog Stereo:playback_F[LR]'</code>. Essentially, these are the names of microphone and speaker devices that the VM would connect to. You can get the names by looking at your patchbay, or run <code>jack_lsp</code>.
#: If there is an existing <code><sound></code> block, it should be removed.
# Finally, we need to set PipeWire's runtime directory and latency. These are configured by environment variables. To do this, the following block right before <code></domain></code>:
#: <syntaxhighlight lang="xml">
  <qemu:commandline>
    <qemu:env name="PIPEWIRE_RUNTIME_DIR" value="/run/user/1000"/>
    <qemu:env name="PIPEWIRE_LATENCY" value="512/48000"/>
  </qemu:commandline>
</syntaxhighlight>
#: If you already have a <code><qemu:commandline></code> section, merge this block with it. Remember to change <code>1000</code> to your actual user ID. You can tune the latency and sampling rate as you like.

If you are using a distro with <code>apparmor</code>, you will need to configure new rules. To do this, append the following lines to your <code>/etc/apparmor.d/local/abstractions/libvirt-qemu</code> (if this file doesn't exist, create it):
<source>
  # PipeWire/JACK
  /etc/pipewire/* r,
  /usr/share/pipewire/* r,
  /run/user/1000/pipewire-0 rw,
</source>

If you are using NixOS, you'll also need to add <code><qemu:env name="LD_LIBRARY_PATH" value="$LD_LIBRARY_PATH:/run/current-system/sw/lib/pipewire"/></code> to your <code><qemu:commandline></code> section.

Once you shutdown your VM and restart it, you should see an emulated HD audio device, and any sound should go into PipeWire. If you don't hear anything, check your patchbay connections.

== Allow QEMU user access to your PipeWire socket ==

We will assuming your username is <code>example</code>, your user ID is 1000, and your virtual machine is called <code>win10</code>. We will also assume the QEMU user is <code>libvirt-qemu</code>, but this varies based on distro. We will be duplicating the socket as <code>/srv/win10/pipewire-0</code>, but the directory can be changed.

First, we must prepare the <code>/srv/win10</code> directory. Run the following commands:

<syntaxhighlight lang="console">
$ sudo mkdir /srv/win10
$ sudo chown example: /srv/win10
$ chmod 700 /srv/win10
$ setfacl -m u:libvirt-qemu:rx /srv/win10
$ touch /srv/win10/pipewire-0
</syntaxhighlight>

We then need to allow the socket to be bind mounted to the new location. To do this, edit <code>/etc/fstab</code>, and add:

 /run/user/1000/pipewire-0 /srv/win10/pipewire-0 none bind,rw,user,noauto 0 0

Then, we configure this socket to be mounted and unmounted when systemd starts <code>pipewire.socket</code>. To do this, run <code>systemctl --user edit pipewire.socket</code> and put in:

<syntaxhighlight lang="ini">
[Socket]
ExecStartPost=/bin/mount /srv/win10/pipewire-0
ExecStopPre=/bin/umount /srv/win10/pipewire-0
</syntaxhighlight>

Then, run <code>systemctl --user restart pipewire.socket</code>, and the socket should be mirrored. Now we can tell libvirt to use this socket.

Run <code>virsh edit win10</code>, and perform the following changes:

# If the <code><domain></code> line doesn't already contain <code>xmlns:qemu</code>, add <code><nowiki>xmlns:qemu='http://libvirt.org/schemas/domain/qemu/1.0'</nowiki></code>. The line should look like:
#: <syntaxhighlight lang="xml"><domain type='kvm' xmlns:qemu='http://libvirt.org/schemas/domain/qemu/1.0'></syntaxhighlight>
# Add the following block to the <code><devices></code> section:
#: <syntaxhighlight lang="xml">
  <audio id="1" type="jack">
    <input clientName="win10" connectPorts="input-name"/>
    <output clientName="win10" connectPorts="output-name"/>
  </audio>
  <sound model="ich9">
    <alias name="sound0"/>
  </sound>
</syntaxhighlight>
#: <code>input-name</code> and <code>output-name</code> should be replaced with a regular expression that matches the name of the JACK input and output ports. Essentially, these are the names of microphone and speaker devices that the VM would connect to. You can get the names by looking at your patchbay, or run <code>jack_lsp</code>.
#: If there is an existing <code><sound></code> block, it should be removed.
# Finally, we need to set PipeWire's runtime directory and latency. These are configured by environment variables. To do this, the following block right before <code></domain></code>:
#: <syntaxhighlight lang="xml">
  <qemu:commandline>
    <qemu:env name="PIPEWIRE_RUNTIME_DIR" value="/srv/win10"/>
    <qemu:env name="PIPEWIRE_LATENCY" value="512/48000"/>
  </qemu:commandline>
</syntaxhighlight>
#: If you already have a <code><qemu:commandline></code> section, merge this block with it. Remember to change <code>1000</code> to your actual user ID. You can tune the latency and sampling rate as you like.

If you are using a distro with <code>apparmor</code>, you will need to configure new rules. To do this, append the following lines to your <code>/etc/apparmor.d/local/abstractions/libvirt-qemu</code> (if this file doesn't exist, create it):
<source>
  # PipeWire/JACK
  /etc/pipewire/* r,
  /usr/share/pipewire/* r,
  /srv/win10/pipewire-0 rw,
</source>

If you are using NixOS, you'll also need to add <code><qemu:env name="LD_LIBRARY_PATH" value="$LD_LIBRARY_PATH:/run/current-system/sw/lib/pipewire"/></code> to your <code><qemu:commandline></code> section.

Once you shutdown your VM and restart it, you should see an emulated HD audio device, and any sound should go into PipeWire. If you don't hear anything, check your patchbay connections.