Microsoft Surface Laptop Ultra (NVIDIA N1X) — hardware support, open issues and testing

[teo-casares]

I've got Omarchy 4 (aarch64, linux-omarchy 7.2.8) running as a daily driver on the Microsoft Surface Laptop Ultra (SKU Surface_Laptop_Ultra_2115, UEFI 7.1.0, NVIDIA N1X, GPU 10de:2e03), dual-booting Windows. As far as I can tell it's the first N1X laptop outside the validated ASUS ProArt P14 / Dell XPS 16, so here's everything it took, in case it helps get this machine supported (cc #15702).

omarchy-hw-platform already reports aarch64-n1x (the Surface exposes NVDA0200), so n1x.sh runs. On top of that, the Surface needs the following.

Status

Area	Status	Fix
USB / I2C / SPI (keyboard, touchpad)	✅	acpi_osi=!
Phantom Volume Down spam	✅ (side buttons lost)	modprobe.blacklist=gpio_keys
Boot from internal disk	✅ via GRUB	Limine resets the firmware before drawing
Touchpad pointer + palm rejection	✅	libinput quirk + hid-multitouch class
NVIDIA GPU (Hyprland, GL 4.6, Vulkan 1.4)	✅	carveout memmap module + small driver patch
Battery / AC	✅	SSAM UART clock + MSHW0733 registry entry
Speakers, headphone jack, mic	✅	SoundWire pad drive + machine-table entry + UCM
Screen brightness, keyboard backlight	✅	small services
Key repeat (doubled letters)	⚠️ worked around	500 ms repeat delay
Webcam	❌	no Linux driver for the "mt8991" camsys/seninf block
USB4 usb4_release=0	disabled, untested
Suspend, external displays	untested

Findings (details + patches available)

1. _OSI("Windows 2022") hides SSPM/USB. The DSDT returns _STA=0 for SVP0, SSPM, HFRP, NH1X and NMAP when Windows 2022 _OSI is true, so mtk_pwrap fails (scmi ret = -19) and xHCI/I2C/SPI fail with -22. acpi_osi=! fixes it; those are the only _OSI uses in the tables.
2. Volume Down (MSHW0040 → soc_button_array → gpio-keys) reports itself held forever. modprobe.blacklist=gpio_keys for now.
3. Limine 12.9.3 resets this firmware (Project Mu) before showing anything, with quiet: no, graphics: no, verbose: yes. It boots fine in QEMU/edk2 with a copy of the ESP. GRUB (the ISO's) works, so I chainload the same UKIs from GRUB on the ESP. Likely suspects: Limine's init_memmap() leaving only 64 MiB below 4 GiB (this machine has ~1.8 GiB there), or disk_create_index()'s recursive ConnectController. I'll report this to Limine separately.
4. NVIDIA: the GPU carveout is missing from the EFI memory map entirely. On the Dell/ASUS it's reserved and has struct pages; on the Surface (efi=debug) the map jumps 0x1fcffffff → 0x300000000, and the IORT RMR gives 0x200000000+4G and 0xa1600000+386M to the GPU. nvidia-drm then hits WARNING … vmap_small_pages_range_noflush / failed to map pages! / Unable to allocate LUT memory and the machine resets. Fix: a tiny module that does memremap_pages(MEMORY_DEVICE_PRIVATE) over those ranges (struct pages, no linear map, same state as the Dell), loaded before nvidia. With it, the stock vmap() path works: 0 warnings, glmark2 5119, vkmark 4282 at 2560x1600, 24 GiB of GPU buffers resident with exact read-back.
5. Battery: 8250_mtk's ACPI path assumes a 26 MHz UART clock for NVDA0240, but it's ~52 MHz, so the SAM link runs at 8 Mbaud and the EC ignores it. Halving the requested baud fixes it (EC fw 6.68.102). Plus surface_aggregator_registry needs an ACPI entry for MSHW0733 (battery + AC only for now).
6. Audio: SoundWire link-0 pads (GPIO66/67) are at 2 mA; ACPI asks for 8 mA, and with 8 mA the bus syncs. The soundwire-mtk machine table also needs an entry for this layout (2x RT1320, with the mic array on the first amp, plus RT721 for the jack). UCM profile included; RT1320 gain capped at -9 dB because there's no speaker-protection data.
7. Touchpad (ELAN 04F3:3358): it advertises pressure 0-2048 but fingers read ~1-40, so libinput rejects every touch (AttrEventCode=-ABS_MT_PRESSURE;-ABS_PRESSURE;, or AttrInputProp=+INPUT_PROP_PRESSUREPAD). Its PTPHQA blob is 4 bytes, so hid-multitouch picks the generic class and drops the Confidence/palm bit; rebinding as MT_CLS_WIN_8 fixes palm rejection.
8. Backlight: no /sys/class/backlight. NVIDIA's firmware brightness query fails and the panel takes nits over eDP DPCD, so I drive it through the RM ioctl interface. The keyboard backlight is a HID feature with no kernel driver.
9. Key repeat: the I2C-HID keyboard sometimes delivers key-release 20-145 ms late in bursts, so it needs a 500 ms repeat delay.

What I'd like to upstream

• An install/hardware/microsoft/fix-surface-laptop-ultra-*.sh set gated by omarchy-hw-aarch64-n1x && omarchy-hw-match "Surface Laptop Ultra": cmdline, libinput quirk, touchpad class, key repeat, UCM.
• Kernel patches for linux-omarchy-aarch64: MSHW0733 registry, 8250_mtk 52 MHz, SoundWire pad drive + machine entry, and the carveout memmap. I'm happy to send these to omarchy-pkgs if that's the right place.
• 10de:2e03 alongside 2e06 where the N1X scripts check the GPU ID.
• The bootloader question: the installer is Limine-only, and Limine doesn't start on this firmware. What's the preferred route?

Full write-up with every file and revert step: https://gist.github.com/teo-casares/8b64949638e1952602a7f435bb34d5b8

[teo-casares]

Update: DAILY DRIVER STATUS Here's the live state from the running machine, Omarchy 4 aarch64 on linux-omarchy 7.2.8-13. I captured it today, 2 h 19 min into the current boot, with 0 failed units and 0 kernel WARNING/Xid/oops lines this boot.

$ cat /sys/class/dmi/id/{sys_vendor,product_name,bios_version}
Microsoft Corporation
Microsoft Surface Laptop Ultra
7.1.0

# GPU: Hyprland on nvidia-drm, panel at native resolution
$ nvidia-smi --query-gpu=name,driver_version,pstate --format=csv,noheader
NVIDIA RTX Spark N1X (6144-core Blackwell RTX GPU), 615.78.08, P8
$ glxinfo -B | grep -E 'renderer|core profile version'
OpenGL renderer string: NVIDIA RTX Spark N1X (6144-core Blackwell RTX GPU)/PCIe
OpenGL core profile version string: 4.6.0 NVIDIA 615.78.08
$ vulkaninfo --summary | grep -E 'apiVersion|deviceType'
	apiVersion         = 1.4.351
	deviceType         = PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
$ hyprctl monitors | grep -A1 Monitor
Monitor eDP-1 (ID 0):
	3270x2180@119.99800 at 0x0
$ dmesg | grep n1x_carveout:
n1x_carveout: memmap for 0x200000000-0x2ffffffff (pfn_valid now 1, refcount 1)
n1x_carveout: memmap for 0xa1600000-0xb7ffffff (pfn_valid now 1, refcount 1)

# Battery / AC over the Surface SAM EC
$ dmesg | grep surface_serial_hub
surface_serial_hub serial0-0: N1X UART clock quirk: requesting 2000000 baud for 4000000 on the wire
surface_serial_hub serial0-0: SAM firmware version: 6.68.102
$ upower -i .../battery_BAT1 | grep -E 'state|percentage|energy-full:'
    state:               fully-charged
    energy-full:         97.12 Wh
    percentage:          100%
$ upower -i .../line_power_ADP1 | grep online
    online:              yes

# Audio over SoundWire (2x RT1320 + RT721)
$ cat /proc/asound/cards | grep -A1 mt8901
 1 [mt8901soundwire]: mt8901-soundwir - mt8901-soundwire
                      MicrosoftCorporation-MicrosoftSurfaceLaptopUltra-124I00142T000M0000000
$ wpctl status   # Built-in Audio
 ├─ Sinks:
 │      Built-in Audio Headphone jack
 │  *   Built-in Audio Speakers
 ├─ Sources:
 │      Built-in Audio Headset microphone (jack)
 │  *   Built-in Audio Internal microphone array

# Screen and keyboard backlight
$ brightnessctl -l | grep Device
Device 'nvidia_0::lcd_backlight' of class 'leds':
Device 'surface::kbd_backlight' of class 'leds':

I've tested the speakers, aux I/O, and internal mic on real workflows and capture. Touch palm rejection is working as well.

Next: I'm testing a Surface Laptop Ultra ISO built on top of the official N1x nightly, in QEMU first and then on hardware. It packages these fixes as DKMS modules plus a support package, and boots without Limine (see Limine-Bootloader/Limine#694). I'll post it here once a fresh install has been verified on the laptop.

Reports filed so far: Limine-Bootloader/Limine#694, NVIDIA/open-gpu-kernel-modules#1458, linux-surface/linux-surface#2307, omacom/omarchy-pkgs#1275, draft PR #15888.

A few corrections to my first post:

• About 1.06 GiB of usable RAM sits below 4 GiB, not ~1.8 GiB.
• power_wrap_drv.usb4_release=0 is the setting that keeps USB4 working on N1x. I haven't enabled or tested it here yet, so USB4 docks don't work yet.
• I had only tested the NVIDIA fix with the memmap module plus my small driver patch. The stock driver with the module alone is still untested.

(Every output above is from this machine.)

[zeroknots]

hey, super helpful. still waiting for delivery of mine. please keep this discussion going.

[teo-casares]

@zeroknots will do, hope yours arrives soon! Here's what has changed since the last update (still linux-omarchy 7.2.8, now -23).

New since last time

• Suspend works. s2idle used to resume to a dead display (nvidia-drm Flip event timeout). options nvidia NVreg_EnableS0ixPowerManagement=1 fixes that: the display comes back on resume. Some resumes still log one Flip event timeout plus a WARN, with no visible effect (reported in N1X (10de:2e03) on Surface Laptop Ultra: GPU carveout missing from EFI memory map, vmap() WARN, LUT allocation fails, oops in nvSetupOutputLUT5, machine resets NVIDIA/open-gpu-kernel-modules#1458).
• No more waking in a bag. With the lid closed in a bag, pressure on the keys woke it every ~30 s (IRQ from the I2C keyboard MSHW0794, found with pm_debug_messages=1). A /usr/lib/systemd/system-sleep/ hook now turns off power/wakeup on the keyboard and touchpad while the lid is closed. It has run on every lid-close suspend since 2026-10-22.
• USB4 / Thunderbolt docks work. Boot with power_wrap_drv.usb4_release=0 and keep the 10de:22cf tunnel root ports out of runtime suspend (a udev rule setting power/control=on). Tested with an OWC Thunderbolt Dock 96W: two 4K60 displays and Ethernet. Caveat: in one boot with the dock attached, every suspend attempt aborted with xhci-hcd NVDA8000:00: PM: dpm_run_callback(): acpi_subsys_suspend returns -22, and logind kept retrying for a day. I haven't isolated that yet, so unplug the dock before suspending.
• Fan: the EC's own fan loop plateaus around 3300 RPM with the SoC at 91 °C under sustained GPU load. Driving the fan from the host through SSAM (FAN, CID 0x0b, an le16 RPM target, as found in linux-surface#1726) works up to ~5700 RPM. I run a small temperature-curve daemon.
• Performance power profile: the EC reads back TMP profile 8 after 4 is set, which the stock surface_platform_profile rejects (EINVAL), so only power-saver/balanced showed up. Accepting 8 as performance gives all three profiles in omarchy powerprofiles.
• Power button works. There is no volume rocker; the phantom "Volume Down" comes from two floating MSHW0040 pins. An hwdb entry maps those two keys to KEY_RESERVED, and gpio_keys is loaded after boot.
• CUDA works (llama.cpp and PyTorch cu130).

A kernel update broke the battery, so the fix no longer patches in-tree drivers
linux-omarchy 7.2.8-23 changed surface_aggregator_registry (the Surface Pro 11 patches) without changing uname -r. My DKMS package carried patched copies of the in-tree SSAM drivers, its guard correctly refused to build them, and battery/AC disappeared until I rebased. The new version uses the stock SSAM drivers unchanged: one small module sets the NVDA0240 UART clock before surface_aggregator opens the port, and another registers the MSHW0733 SSAM clients (battery, AC, platform profile, sensors, fan). It has been loaded on every boot since 2026-10-24.

The UART clock turned up a real 8250_mtk bug: mtk8250_set_termios() computes uartclk * 100 in 32 bits, which wraps for any clock above 42.9 MHz. At 52 MHz and 4 Mbaud it programs the wrong fractional divider, so the 52 MHz patch in my gist was not right on its own. A fix for linux-serial is ready to send.

Corrections to my first post

• Touchpad: the PTPHQA blob is not 4 bytes. It's the full 256 bytes, but it's declared on the Digitizers usage page (0x000d00c5) instead of the vendor page (0xff0000c5), so hid-core doesn't classify the pad as a Win8 PTP. A hid-multitouch device-table entry for 04F3:3358 (MT_CLS_WIN_8) fixes palm rejection; that patch is ready to send to linux-input.
• NVIDIA: still only tested with the memmap module plus my driver patch. NVIDIA has picked up the issue (open-gpu-kernel-modules#1458, internal bug 6956987).

Still open: the webcam (no driver for the camera block), suspend with the dock attached, and Limine on this firmware (a fix is in testing in Limine-Bootloader/Limine#694).

The gist still has the 2026-10-18 files. I'll update it and link it here once the new versions are cleaned up.

[teo-casares]

@zeroknots will do, hope yours arrives soon! Here's what has changed since the last update (still linux-omarchy 7.2.8, now -23).

New since last time

• Suspend works. s2idle used to resume to a dead display (nvidia-drm Flip event timeout). options nvidia NVreg_EnableS0ixPowerManagement=1 fixes that: the display comes back on resume. Some resumes still log one Flip event timeout plus a WARN, with no visible effect (reported in N1X (10de:2e03) on Surface Laptop Ultra: GPU carveout missing from EFI memory map, vmap() WARN, LUT allocation fails, oops in nvSetupOutputLUT5, machine resets NVIDIA/open-gpu-kernel-modules#1458).
• No more waking in a bag. With the lid closed in a bag, pressure on the keys woke it every ~30 s (IRQ from the I2C keyboard MSHW0794, found with pm_debug_messages=1). A /usr/lib/systemd/system-sleep/ hook now turns off power/wakeup on the keyboard and touchpad while the lid is closed. It has run on every lid-close suspend since 2026-10-22.
• USB4 / Thunderbolt docks work. Boot with power_wrap_drv.usb4_release=0 and keep the 10de:22cf tunnel root ports out of runtime suspend (a udev rule setting power/control=on). Tested with an OWC Thunderbolt Dock 96W: two 4K60 displays and Ethernet. Caveat: in one boot with the dock attached, every suspend attempt aborted with xhci-hcd NVDA8000:00: PM: dpm_run_callback(): acpi_subsys_suspend returns -22, and logind kept retrying for a day. I haven't isolated that yet, so unplug the dock before suspending.
• Fan: the EC's own fan loop plateaus around 3300 RPM with the SoC at 91 °C under sustained GPU load. Driving the fan from the host through SSAM (FAN, CID 0x0b, an le16 RPM target, as found in linux-surface#1726) works up to ~5700 RPM. I run a small temperature-curve daemon.
• Performance power profile: the EC reads back TMP profile 8 after 4 is set, which the stock surface_platform_profile rejects (EINVAL), so only power-saver/balanced showed up. Accepting 8 as performance gives all three profiles in omarchy powerprofiles.
• Power button works. There is no volume rocker; the phantom "Volume Down" comes from two floating MSHW0040 pins. An hwdb entry maps those two keys to KEY_RESERVED, and gpio_keys is loaded after boot.
• CUDA works (llama.cpp and PyTorch cu130).

A kernel update broke the battery, so the fix no longer patches in-tree drivers
linux-omarchy 7.2.8-23 changed surface_aggregator_registry (the Surface Pro 11 patches) without changing uname -r. My DKMS package carried patched copies of the in-tree SSAM drivers, its guard correctly refused to build them, and battery/AC disappeared until I rebased. The new version uses the stock SSAM drivers unchanged: one small module sets the NVDA0240 UART clock before surface_aggregator opens the port, and another registers the MSHW0733 SSAM clients (battery, AC, platform profile, sensors, fan). It has been loaded on every boot since 2026-10-24.

The UART clock turned up a real 8250_mtk bug: mtk8250_set_termios() computes uartclk * 100 in 32 bits, which wraps for any clock above 42.9 MHz. At 52 MHz and 4 Mbaud it programs the wrong fractional divider, so the 52 MHz patch in my gist was not right on its own. A fix for linux-serial is ready to send.

Corrections to my first post

• Touchpad: the PTPHQA blob is not 4 bytes. It's the full 256 bytes, but it's declared on the Digitizers usage page (0x000d00c5) instead of the vendor page (0xff0000c5), so hid-core doesn't classify the pad as a Win8 PTP. A hid-multitouch device-table entry for 04F3:3358 (MT_CLS_WIN_8) fixes palm rejection; that patch is ready to send to linux-input.
• NVIDIA: still only tested with the memmap module plus my driver patch. NVIDIA has picked up the issue (open-gpu-kernel-modules#1458, internal bug 6956987).

Still open: the webcam (no driver for the camera block), suspend with the dock attached, and Limine on this firmware (a fix is in testing in Limine-Bootloader/Limine#694).

The gist still has the 2026-10-18 files. I'll update it and link it here once the new versions are cleaned up.

[ikegam]

Thank you for your hard work. I received one as well, and it works great with your setup. I’m now working on the camera with Claude, and it’s starting to give data. I think it will take a few more days to get the full set of drivers ready.