Skip to content

Support the NVIDIA N1x laptops - #14820

Merged
ryanrhughes merged 51 commits into
quattrofrom
n1x
Oct 15, 2026
Merged

ryanrhughes merged 51 commits into
quattrofrom
n1x

Conversation

@ryanrhughes

@ryanrhughes ryanrhughes commented Oct 10, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

Runtime support for the NVIDIA N1x laptops (ASUS ProArt P14 H7407BA, Dell XPS 16 DX16263): ACPI-only aarch64 machines with a MediaTek MT8901 CPU complex and an NVIDIA GB20B GPU, running linux-omarchy for aarch64 and NVIDIA's open driver (omacom/omarchy-pkgs#930, now on edge). Also brings the Snapdragon and NVIDIA GB10 platform work that landed on this branch. Up to date with today's quattro.

Platforms

N1x hardware setup (install/hardware/n1x.sh and model scripts, with migrations for existing installs)

  • The kernel comes from the installer and the N1x package list. The GPU drives the panel on release firmware (early KMS, falling back to the firmware framebuffer when DKMS did not build), with Plymouth through LUKS unlock and a fallback boot entry.
  • The firmware reserves 4 GiB for the GPU through NVIDIA's EXPOSED_DISP_FRM variable, instead of the kernel reclaiming the 62.5 GiB it set aside for Windows. Swap on zram, memory tuning, rtkit.
  • USB4/Thunderbolt: host routers kept powered, bus numbers set aside for docks, tunnel root ports kept in D0 so a dock's PCIe devices appear. Docks are approved through quattro's Thunderbolt authorization like any other accessory.
  • CPPC autonomous selection turned on through a udev rule that first restores each policy's full range, so power-profiles-daemon's energy preference takes effect.
  • Audio: NVIDIA's MT8901 SoundWire UCM profiles from alsa-ucm-conf#862, unchanged, for both models until alsa-ucm-conf ships them. ProArt P14: ASUS's amplifier tuning and a speaker tuning. The Dell loses its headset microphone, which NVIDIA does not declare on that board.
  • ProArt P14: touchpad quirk, media keys on the top row with F1-F12 on Fn+Esc.
  • The lid no longer suspends a laptop driving a display whose connector type is Unknown.
  • systemd 262's NvPCR units no longer fail on every boot.

Fixes that are not N1x-specific

  • arp_ignore=1 / arp_announce=2, so a dock's Ethernet and Wi-Fi on one network don't answer for each other and make NetworkManager drop the wired routes.
  • Arch Linux ARM's signing key is repaired when pacman's prompt left it present but untrusted. This now sits alongside quattro's --needed keyring refresh.
  • Gliff installs on aarch64 now that edge builds it there; NVIDIA's 32-bit packages and lib32 drivers are skipped where multilib doesn't exist (Fix NVIDIA package selection on ARM #10752); omarchy-pkg-drop runs as root like omarchy-pkg-add.

Test plan

  • ./test/cli passes. ./test/shell fails only the 15 cases that fail identically on plain quattro on this machine (missing sibling checkouts, root-owned /usr paths, session environment).
  • Tests for platform detection (N1x by ACPI or GPU, GB10, Snapdragon, contradictions), package lists, pacman templates, image manifests, device-tree matching and keyring repair.
  • An N1x ISO built from this branch (omacom/omarchy-iso n1x) installs unattended in a KVM VM and reaches Hyprland with no failed units.
  • Running on a ProArt P14: detected as aarch64-n1x, suspend at about 0.77 W with a clean resume, a CalDigit TS4 driving two monitors through unplug and replug, power profiles switching EPP, and speakers on NVIDIA's profiles.

spencerbull and others added 29 commits October 12, 2026 01:50
Port of the N1x bring-up onto quattro. Everything is gated on the new
omarchy-hw-n1x detector (ACPI NVDA0200 MediaTek I2C controllers or the
10de:2e06 GB20B GPU, aarch64 only), so x86_64 behaviour is unchanged.

- install/hardware/n1x.sh (runs before nvidia.sh): ensures linux-n1x and
  headers, drops the stock kernel, pins console=tty0 acpi=nospcr and a
  verbose boot through an alphabetically-first Limine drop-in (the entry
  tool prepends += fragments, so that drop-in's fragment lands last and
  wins over omarchy-defaults.conf's quiet loglevel=0), adds a
  linux-n1x-rescue entry with its own KERNEL_CMDLINE key and UKI (Limine
  hands the entry cmdline to systemd-stub as load options, overriding the
  embedded one), keeps the NVIDIA stack unloaded with blacklist plus
  install-to-/bin/false lines because 610.57.04 cannot boot this GPU's GSP
  and its dead render node aborts Hyprland, and sets up key-only SSH for
  the install user plus a per-boot hardware probe.
- install/hardware/nvidia.sh defers to n1x.sh on this platform instead of
  configuring early KMS.
- etc/mkinitcpio.conf.d/omarchy_hooks.conf drops plymouth (crashes in the
  AArch64 initramfs) and microcode on aarch64 and adds the I2C-HID input
  modules the internal keyboard and touchpad need before root is mounted.
- Arch Linux ARM pacman config for installed aarch64 systems; the bundled
  Node.js tarball lookup follows the machine architecture.
- test/shell.d/hw-n1x-test.sh covers the detector and the script contract.
BIOS 1.0.4 lets the 610.57.04 open driver boot the GPU's GSP, so the
NVIDIA blacklist is now limited to pre-release 0.x firmware. On 1.0+ the
script installs the driver, enables nvidia_drm modeset with fbdev, loads
the modules in the initramfs for early KMS, and blacklists the firmware
framebuffer driver on the kernel command line, as NVIDIA does on the
Spark, so Hyprland sees a single DRM device. Verified on the dev unit:
eDP-1 at 1920x1200@120 after an unattended reboot.
…'s aarch64 repo

- hardware/n1x.sh only enables early KMS when DKMS produced nvidia_drm for the
  N1x kernel; otherwise the NVIDIA stack stays unloaded and Hyprland renders on
  the firmware framebuffer instead of every UKI build failing.
- Remote access moves out of the runtime: the N1x development ISO owns its
  own, separately removable SSH setup.
- Installed aarch64 systems gain pkgs.omarchy.org's edge aarch64 repository,
  ahead of Arch Linux ARM, and omarchy-refresh-pacman no longer copies an
  x86_64 channel config onto them.
…e N1x probe

The DKMS check now requires all four modules the initramfs lists, not just
nvidia_drm. The probe reported a recovery key by name on every boot whether or
not sshd was set up; it now prints what systemd says about sshd.
It is a text console with the GPU off; SSH comes only from a --dev-ssh image.
Its PixArt haptic pad reports force on a 0-65535 pressure axis, so
libinput's default touch threshold is never reached. Mark it a pressure
pad in /etc/libinput/local-overrides.quirks, the only file libinput reads
from /etc, and add a migration for existing installs.
The MediaTek MT8901 SoundWire card (cs42l43 + 2x cs35l56) exposes only
front-end PCMs; without a UCM profile nothing routes them to the codecs
and PipeWire offers a dummy output. Add a profile for this model that
sets the AFE interconnect and reuses alsa-ucm-conf's cs42l43/cs35l56
controls. It lives under its own Omarchy UCM directory, selected through
a conf.d link for the card's long name, for omarchy-settings to install
under /usr/share/alsa/ucm2. A migration restarts PipeWire to pick it up.
pacman-aarch64.conf puts [omarchy] ahead of Arch Linux ARM so Omarchy's
builds, the NVIDIA driver among them, win. Installs from earlier N1x
images list it last. Move the section in place, leaving the rest of the
file as it is.
…down

The N1x console drop-in added loglevel=7, systemd.show_status=1 and
friends for bring-up, after Omarchy's quiet splash so they would win.
Kernel and systemd messages then printed over Plymouth once the disk
was unlocked and again during shutdown. Keep only console=tty0 and
acpi=nospcr on the normal entry; the rescue entry stays verbose.
The panel stays at its dim power-on backlight until the NVIDIA backlight
is first written: its DPCD brightness registers read zero while nvidia_0
reports 100. systemd-backlight is the first to write it, after the disk
is unlocked, so the LUKS prompt was nearly black. An initramfs hook right
after plymouth sets 60%; the saved level still returns after unlock.
The 128 GB ProArt P14's firmware reserves 62.5 GiB as the N1x GPU's
dedicated memory under Windows. Linux's NVIDIA driver runs the GPU from
system memory, so the reservation sits idle and the laptop boots with
61 GiB, and the BIOS has no setting to shrink it.

linux-omarchy-n1x's efi_reclaim_reserved= hands the idle 61.9 GiB to
the kernel, leaving reserved the firmware's DRAM training buffer and
the top 512 MiB, where the GPU's firmware runs. Only add it when the
firmware still reserves exactly the block it was measured on.
On Windows, ASUS's Dolby Atmos layer voices this laptop's speakers on
top of the Cirrus amplifier tuning; on Linux they sounded muffled. Start
from the static speaker voicing in ASUS's Dolby tuning for this model
(its 600 Hz speaker PEQ and 20-band optimizer curve, fitted to 0.06 dB),
then voice the low mids by ear in place of Dolby's adaptive dialog
enhancement: a 300 Hz mud cut, a softer 600 Hz bump and a broad 2.5 kHz
vocal lift. Above 4 kHz it is Dolby's curve.

It applies only once the amplifiers have ASUS's tuning for this laptop,
which the curve assumes.
The amplifiers ran their ROM defaults, quiet and unvoiced, without
ASUS's tuning for this laptop. Install asus-proart-p14-speaker-firmware
on the H7407BA, ahead of the speaker tuning that builds on it, and add
both to existing installs.
- Install linux-omarchy-n1x, which carries the N1x platform, USB4 and ASUS
  ProArt P14 patches, and drop the stock and linux-n1x kernels.
- Let limine-mkinitcpio-hook build the rescue entry as the kernel's fallback
  UKI, so it follows kernel updates, instead of a one-off copied UKI.
- Keep Plymouth in the initramfs on aarch64 again; only the N1x gets the
  I2C keyboard modules for the LUKS prompt.
- Suspend to idle, as the firmware's deep sleep returns at once.
- Keep the USB4 host routers powered and leave PCI room for docks on the
  ProArt P14, where USB4 is validated.
- Hold the GPU stack back only on Dell's pre-release firmware; the version
  sort let ASUS firmware through by accident.
- Run omarchy-n1x-probe on demand instead of on every boot.
Installs the kernel, writes the same keyboard, sleep, USB4 and rescue-entry
drop-ins as a fresh install, removes the one-off rescue UKI and the
every-boot probe, and drops linux-n1x once the machine no longer runs it.
The boot order only changes once DKMS has built the NVIDIA modules for the
new kernel.
The installer now gives every aarch64 install console=tty0 acpi=nospcr,
the setting the N1x's serial console table needs, so n1x.sh no longer
writes its own copy of it.
Arch Linux ARM signs every package in its repositories, archlinux-keyring
included, with its build key 68B3537F39A313B3E574D06777193F152BDBE6A6, and
only archlinuxarm-keyring carries it. pacman's "Import PGP key?" prompt adds
the key without trusting it, after which every update stopped at "unknown
trust" with nothing to repair it.

Before the keyring refresh, an aarch64 machine whose key is not valid
populates it from the installed archlinuxarm-keyring, or fetches it by
fingerprint, signs it locally and installs the package when that is missing,
the way the Omarchy key is bootstrapped. keys.openpgp.org does not carry it,
so it comes from keyserver.ubuntu.com. After the refresh, a key that is still
untrusted fails the update instead of reporting the keys correct.
The ProArt P14's USB4 tunnel root ports (10de:22cf) cannot signal a hotplug
while runtime-suspended, in D3hot or D3cold alike. They suspend about 100 ms
after going idle, so by the time a dock is plugged in and bolt authorizes it,
the PCIe tunnel comes up behind a sleeping port and nothing is enumerated: the
dock shows as authorized, its USB and display work, and its Ethernet never
appears until something happens to read the port's config space. On a fresh
install from the n1x ISO, two of three physical replugs of a CalDigit TS4
left it without its NIC, one of them for 97 minutes. Holding the ports in D0
brought the NIC up 16 ms after authorization on every try, and a deauthorize
and authorize cycle went from no NIC after 30 s to the NIC in 1 s.

Keep the three ports in D0 with a udev rule, written alongside the USB4
command line on the ProArt, and add it to existing ProArt installs with a
migration that also applies it to the running system. linux-omarchy-n1x
should eventually resume the root port when it activates a PCIe tunnel
instead; this costs the idle power of three root ports until then.
With the kernel's default ARP behaviour every interface answers for every
local address. Plug a dock's Ethernet into the same network the laptop's
Wi-Fi is on, and the Wi-Fi answers NetworkManager's address-conflict probe
for the new wired address: the probe reaches the Wi-Fi a few hundred
milliseconds late over the air, after the address is already up. NetworkManager
then sees "conflict detected ... with host <the laptop's own Wi-Fi MAC>" and
withdraws the wired routes while still reporting the connection as connected,
so all traffic quietly stays on Wi-Fi. It is a race, so it only happens on
some activations. LAN hosts also learn the Wi-Fi address on the wired MAC
and the other way round. A fresh N1x install docked to a CalDigit TS4 lost
its wired routes on two of its first three activations this way.

Set arp_ignore=1 and arp_announce=2 for all and default interfaces: answer
only for the interface's own address and announce from it. With that, the
late probes still reach the Wi-Fi but go unanswered, and three further
activations and a replug saw no conflict. The settings go in their own file
because an older source's aarch64 omarchy-settings-dev replaces
99-omarchy-sysctl.conf with a single line. A migration applies them to the
running system.
The rescue entry carries its own command line and missed the
initramfs_async=0 every other entry gets.

[Ryan: cherry-picked from 3b93764 onto n1x. Its other half, removing only
the stray kernels that are installed, is already how n1x removes them
(omarchy-pkg-drop skips what isn't installed). The migration that moves
existing N1x installs writes the same rescue line, so it gets the flag too.]
…uling

omarchy-settings ships the x86_64 memory stack on aarch64 now (swap on zram,
zswap off, the reclaim sysctls, systemd-oomd's slice policy), but installs
from the earlier N1x images ran with no swap at all. They also lacked rtkit,
which nothing pulls in on aarch64, so PipeWire got no realtime scheduling.

rtkit joins the aarch64 package list, and a migration gives existing N1x
installs rtkit and zram-generator and starts the zram swap without a reboot.
…red and set aside bus numbers for docks

[Ryan: cherry-picked onto n1x, which already sets the ASUS ProArt P14 up with
the larger windows and root-port rule validated on it (Spencer: "take
yours"). This setting becomes the one for every other N1x, the XPS 16 among
them, and its migration leaves the ProArt to 1791077437.]
…d RAM, since a large external display reaches into it
The n1x-xps16 branch built the N1x rescue entry as its own UKI and rebuilt it
from a pacman hook running omarchy-refresh-n1x-rescue. On n1x the
limine-mkinitcpio-hook fallback entry does that job and the command does not
exist, so an install that ran that branch would fail the hook on every kernel
update and keep a rescue entry for an old kernel. Remove both there.
…rm64 client and multilib does not exist

omarchy-install-gaming-gpu-lib32 asked for lib32-nvidia-utils on the N1x, which no aarch64 repository has, so omarchy-install-gaming-steam (and the Heroic, Lutris and Battle.net installers) stopped before installing anything. omarchy-pkgs now builds steam for aarch64 on Valve's native arm64 client (omacom/omarchy-pkgs#842), which needs no lib32 drivers, so the same installer works on both architectures.
emirb and others added 4 commits October 13, 2026 23:09
omarchy-pkgs renames linux-omarchy-n1x to linux-omarchy (omarchy-pkgs
#1038): the aarch64 kernel takes the name the x86_64 one has.

The N1x package list and its hardware setup ask for it by that name
now: the packages to install, the fallback image and boot order Limine
is given, and the check that one such kernel is installed.

The migrations that moved N1x laptops onto linux-omarchy-n1x are left
as they were. The laptops that ran them are being reinstalled, and a
fresh install runs none of them.
The hardware setup installed linux-omarchy and its headers itself, and
kept NVIDIA's older linux-n1x in place until the machine had booted the
new kernel. Both were for laptops moving over from linux-n1x, and those
are being reinstalled.

On an install from the ISO neither does anything: the installer puts the
kernel and its headers on the machine and checks they match, and the
N1x package list names them. The list is now the one place in this
repository that names the N1x's kernel.

Other kernels are still dropped, so Limine shows one.
Install the N1x's kernel as linux-omarchy
…effect

cppc_cpufreq leaves autonomous selection off, so the CPUs follow the
kernel governor and the energy performance preference power-profiles-
daemon sets per profile changes nothing. NVIDIA's RTX Spark driver for the
daemon relies on it being on, as their images set it. The firmware hands
over each core with its CPPC floor raised to its ceiling for a faster
boot, which autonomous selection would keep, so the cores would never
slow down.

As NVIDIA's power-knobs package does, a udev rule restores each policy's
full range when cppc_cpufreq loads and when a CPU comes online, then turns
autonomous selection on, which programs the floor and ceiling from that
range. The setup writes it on new installs and a migration on existing
N1x machines; it takes effect at the next boot.
These are NVIDIA's profiles from alsa-ucm-conf#862 (822569010c),
unchanged. They pick a profile per SoundWire codec topology instead of
per machine, and cover the ProArt P14 and the Dell XPS 16 along with
NVIDIA's other RTX Spark boards. The Dell loses the headset microphone,
which NVIDIA does not declare on that board.

The files move again when alsa-ucm-conf merges #862 under the card's
upstream name, mt8901-sdw.
# Conflicts:
#	bin/omarchy-setup-direct-boot
#	bin/omarchy-update-keyring
#	default/omarchy/omarchy-menu.jsonc
#	install/omarchy-x86_64-only.packages
#	test/shell.d/update-keyring-test.sh
@ryanrhughes
ryanrhughes marked this pull request as ready for review October 15, 2026 05:08
@greptile-apps

greptile-apps Bot commented Oct 15, 2026 •

Copy link
Copy Markdown
Contributor

RetriggerView in GreptileConfidence Score: 5/5

[High impact] The PR appears safe to merge, with a non-blocking retry gap remaining in the firmware memory migration.

Findings

  1. P2 Retry skips unfinished boot rebuild ▶

Summary

Adds NVIDIA N1x laptop support and separate package sets for N1x, GB10, and Snapdragon machines.

  • Omarchy names N1x, GB10, and Snapdragon as separate ARM platforms.
  • N1x laptops get a dedicated kernel, boot path, and hardware setup.
  • N1x firmware reserves 4 GiB for the GPU instead of 62.5 GiB.
  • The ProArt P14 gets tuned speakers, working controls, and its own key layout.
  • Snapdragon laptops get the boot and firmware setup their hardware needs.
  • GB10 systems skip unsupported sleep and use the boot path suited to the DGX Spark.
  • Shared setup fixes cover ARM packages, signing keys, and network interfaces.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  Detect[Detect hardware platform] --> N1x[NVIDIA N1x]
  Detect --> GB10[NVIDIA GB10]
  Detect --> Qualcomm[Snapdragon]
  N1x --> Laptop[Configure graphics, input, audio, power and USB4]
  GB10 --> Desktop[Disable sleep and configure DGX Spark boot]
  Qualcomm --> Drivers[Configure device trees, drivers and firmware]
  Laptop --> Boot[Rebuild boot files]
  Desktop --> Boot
  Drivers --> Boot
Loading

Reviews (2) · Last reviewed commit: "Retry the N1x boot rebuild until the rec..." · Reviewed by Greptile

Comment thread migrations/1791542175.sh Resolved
Comment thread migrations/1791784278.sh Outdated
Comment on lines +12 to +14
if [[ -f $dropin ]]; then
sudo rm "$dropin"
sudo limine-mkinitcpio

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Retry skips unfinished boot rebuild

If limine-mkinitcpio fails after this block deletes the drop-in, retrying the migration skips the rebuild because the file is already gone. The retry then succeeds and gets marked complete while the boot files can still contain efi_reclaim_reserved=. Keep a pending-rebuild marker, or otherwise retry the rebuild until it succeeds.

Knowledge Base Used: Migration framework

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed in 3af3648: the migration also rebuilds while the running kernel's command line still has efi_reclaim_reserved=, so a retry after a failed rebuild rebuilds again. Covered by test/shell.d/n1x-gpu-memory-firmware-migration-test.sh.

@ryanrhughes
ryanrhughes merged commit 246f33a into quattro Oct 15, 2026
2 checks passed
omarchybot added a commit to fldc/omarchy that referenced this pull request Oct 15, 2026
…otstrap

quattro now bootstraps and checks Arch Linux ARM's signing key before the keyring refresh (omacom#14820), and its new test cases assert the unpinned refresh command. Strip sudo env in the stub before the new gpg/pacman-key dispatch, and expect LC_ALL=C in the two new refresh assertions.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
omarchybot added a commit to Arthur031221/omarchy that referenced this pull request Oct 15, 2026
quattro gated fix-surface-keyboard.sh on x86_64 and made its config directory overridable (omacom#14820). Keep both, and write the AMD fallback through the same directory. The migration skips non-x86 machines too, since the script now writes nothing there and the rebuild would be wasted. surface-hardware-test.sh asserted the old behaviour (nothing written without pinctrl, MODULES= rather than MODULES+=), and the PR's own tests now stub omarchy-hw-x86.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@emirb emirb added this to the 4.5 milestone Oct 17, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

7 participants