voidhw

Hardware-aware driver and firmware selection for Void Linux.

voidhw reads the machine's PCI and USB devices, CPU vendor and chassis type from sysfs, and works out which packages, config files and init services it needs: Mesa for Intel and AMD GPUs, the right NVIDIA driver branch for each NVIDIA GPU, Wi-Fi firmware and DKMS drivers, CPU microcode, laptop power management, hybrid-graphics switching, and guest tools when it runs in a VM (VMware, VirtualBox, QEMU/KVM; detected from DMI strings and the virtual PCI devices, as systemd-detect-virt does).

By default it only reports. Nothing changes unless you pass --apply.

$ voidhw
Machine: notebook (CPU GenuineIntel)

Devices:
  0000:00:14.3 8086:a370 -> wifi-intel-firmware
  0000:01:00.0 GPU 10de:2191 -> nvidia-current

Repositories: nonfree
Packages:     linux-firmware-intel nvidia intel-ucode power-profiles-daemon
File:         /etc/modprobe.d/voidhw-nvidia-drm.conf
Services:     power-profiles-daemon
...

Usage

voidhw                       report what this machine needs
voidhw --json                the same plan as JSON (for installers)
voidhw --apply --dry-run     preview the install steps
sudo voidhw --apply          install packages, write files, enable services
voidhw --root /mnt/target    inspect or (with --apply) configure a system mounted elsewhere
voidhw --root /mnt/target --hardware-from /   configure the target for the machine running this
voidhw --profiles DIR        add or override profiles from *.toml files
voidhw --snapshot DIR        save the hardware description (no serial numbers) for a bug report

Services are enabled for the init system found under --root (dinit or runit; override with --init); a service the init has no definition for is skipped with a message.

NVIDIA branches

Void ships four NVIDIA driver branches (nvidia, nvidia580, nvidia470, nvidia390). The branch for each GPU comes from NVIDIA's own supported-chips lists, compiled into data/nvidia-branches.toml by tools/gen-nvidia-table.py: the newest branch that lists the PCI device id wins. GPUs no Void driver supports stay on nouveau. Regenerate the table when Void moves to a new branch.

Profiles

Profiles are TOML in data/profiles/ (compiled in) or a directory given with --profiles. See src/profile.rs for the schema. tools/check-packages.py verifies that every package exists in the Void repositories and that every service is shipped by one of the profile's packages.

Testing

cargo test runs the unit tests and integration tests against sysfs trees: real snapshots in tests/fixtures/ and synthetic machines built in the tests. To add your machine: voidhw --snapshot tests/fixtures/<name> and write a test for the expected plan.

Lineage

The profile idea comes from CachyOS's chwd, via the chwd_port prototype in Aetheris OS (GPL-3.0). This is a rewrite for Void: the PCI class parsing is fixed, package names are Void's, and changes are opt-in.

Licensed under GPL-3.0-or-later.