Hardware and Driver Manager (linxira-hardware-driver-manager)
Why it exists
Detection answers “what hardware is present”. It does not answer “which driver, which packages, and how does a failure roll back”. Without that separation the user sees a recommendation but not the plan behind it.
This manager turns hardware facts into reviewable fixed-policy plans, and hands execution to the single system transaction backend.
Responsibilities
It owns:
- Reading the
linxira-hwd-detectorJSON with strict schema, name, version and duplicate-key validation. - A fixed policy table: Intel / AMD Mesa, NVIDIA open / proprietary, and Hyper-V / QEMU / VMware / VirtualBox guests.
- Installed-state reporting from fixed evidence only:
linux,linux-ltsand their headers, DKMS, initramfs, loaded modules and the reboot marker. - Atomically saved plans (
0600) under$XDG_STATE_HOME/linxira/hardware-driver-manager/plans. - A root-owned read-only diagnosis, and confirmed installation transactions for three VM guest policies.
It does not own:
- Hardware detection (Hardware Detection).
- The transaction backend, snapshots or Polkit (Components).
- Kernel packages themselves (Kernel Manager).
Chain of responsibility: who reads, who plans, who changes
| Step | Performer | Changes the system? |
|---|---|---|
| Collect hardware facts | Hardware Detection (no-argument JSON) | Read-only |
| Facts → fixed policy → plan | This manager | Read-only; plans go to the user state directory |
| Driver installation, snapshot, service verification | Components system transaction service + Polkit | Privileged, with confirmation and a Timeshift snapshot |
Every local action stops at a plan. Apply is not executed by this program: it asks
org.linxira.Components1 for a plan and for confirmation to run it; the backend performs
the package installation, service enablement and snapshot.
How to use
GUI
Launch Linxira Hardware and Driver Manager from the application menu, or run:
linxira-hardware-driver-manager
Four tabs:
| Tab | Contents |
|---|---|
| Overview | Detected hardware class IDs, detector name and version, Run root diagnosis |
| Drivers | Recommended policy, full plan JSON, Save plan and Apply |
| Kernel compatibility | linux / linux-lts kernel, headers and DKMS table |
| Activity / report | Complete read-only report JSON |
The Drivers tab lists only policies whose hardware class was detected; with no match the
list is empty. Saving needs the I reviewed this exact plan checkbox. Apply needs the
same checkbox plus a policy from the three VM guest set and an applicable plan; otherwise
it stays disabled, and for other policies it shows
Only applicable reviewed VM guest policies have an executable backend.
Command line
linxira-hardware-driver-manager --report-json
linxira-hardware-driver-manager --plan org.linxira.driver.intel-mesa.v1
linxira-hardware-driver-manager --plan org.linxira.driver.intel-mesa.v1 --json
--plan accepts only the fixed policy IDs below; no package name, path or command. Plans
are always written under $XDG_STATE_HOME/linxira/hardware-driver-manager/plans/
(~/.local/state/... when unset), named after the plan's own SHA-256. Without --json the
command prints the plan path; with --json it prints the plan. Exit status is 0 when the
plan is applicable and 3 otherwise; when the detector is unavailable both --report-json
and --plan exit with status 2.
| Policy ID | Hardware fact | Current status |
|---|---|---|
org.linxira.driver.intel-mesa.v1 |
graphics.intel |
Plan only; Apply is not offered |
org.linxira.driver.amd-mesa.v1 |
graphics.amd |
Plan only; Apply is not offered |
org.linxira.driver.nvidia-open-dkms.v1 |
graphics.nvidia |
Plan only; explicit review required |
org.linxira.driver.nvidia-proprietary-dkms.v1 |
graphics.nvidia |
Unavailable: no dual-kernel proprietary NVIDIA DKMS package in the reviewed Arch cohort |
org.linxira.driver.vm-qemu-guest.v1 |
vm.qemu |
Apply available in the GUI |
org.linxira.driver.vm-virtualbox-guest.v1 |
vm.virtualbox |
Unavailable: dual-kernel guest-module providers are not fixed |
org.linxira.driver.vm-vmware-guest.v1 |
vm.vmware |
Apply available in the GUI |
org.linxira.driver.vm-hyperv-guest.v1 |
vm.hyperv |
Apply available in the GUI |
A plan records the policy ID, the hard-coded package set, the detector digest, licence and driver model, kernel/header and DKMS preconditions, expected effects, warnings, blockers, applicability and the plan's own SHA-256. Plans are written atomically with private permissions; a plan directory that is itself a symlink is rejected.
Transaction boundary
- A local plan is a review artifact, not an execution token. Plans from the CLI and the
GUI preview describe what would happen; the CLI has no apply subcommand, and a local
plan's
applyfield is always{"implemented": false, "result": "backend-not-ready"}. - Read-only diagnosis goes through the system D-Bus service
org.linxira.Components1with the fixed operationorg.linxira.hardware.driver-state-diagnose.v1, an empty parameter object and a plan whose risk isread-only. After the user confirms in the transaction service, the receipt must reportstatus = succeededandchanged = false. Nothing in this step touches packages, services or system files. - Apply is available only for the three VM guest policies (Hyper-V, QEMU, VMware). On
Apply the manager first asks the transaction service for a root-owned, reviewable driver
plan whose risk is
system-change-reboot-possibleand whose rollback is the pre-change Timeshift snapshot; after confirmation the backend creates the snapshot, applies the fixed package set, verifies platform services and returns an immutable receipt. A change in repository or hardware evidence between the two steps invalidates the plan. - VirtualBox, proprietary NVIDIA and the ordinary Intel / AMD / NVIDIA graphics plans have
no executable backend; their Apply control stays disabled with
Apply unavailable: backend-not-ready.
When root is required
- Not required:
--report-json,--plan, browsing, policy preview and plan saving. - Required:
Run root diagnosisand applying a VM guest policy. Both go through the system transaction service, where Polkit decides authorisation and prompts when needed; the manager never callssudo,pkexecor a package manager itself.
Known limitations
- The support matrix is
linuxandlinux-ltsplus their headers. DKMS-based policies report a blocker when a required kernel is missing. - NVIDIA open and proprietary are alternatives requiring an explicit choice. Device-ID support is not claimed; the proprietary DKMS comparison is blocked.
- Installed state comes only from fixed evidence:
uname -r,/usr/lib/modules,pacman -Q,dkms status, initramfs under/boot,/proc/modules,/proc/uptimeand/run/reboot-required; no arbitrary path is scanned. - Detector output is capped at 1 MiB with a 10-second timeout. Without the detector the report is unavailable (CLI exit status 2).
- Plan directories and files reject symlinks and use
0600file permissions. - Internal subprocesses use fixed argument lists,
shell=Falseand an 8-second timeout; no path, package name or command input is accepted.
Design notes
Current behaviour and boundaries are in the
linxira-hardware-driver-manager README,
maintainer conventions in AGENTS.md. The transaction shape is described in
Components.