watchthelight labs

Upstream scout runs: 2026-10-27

Second run (Mac, 2026-10-27 09:29 to 18:11 UTC)

All eleven targets. 323 API calls to collect, 0 errors, no rate limit. Three web sweeps (xe, nvgfx, Wi-Fi and Bluetooth) ran before the session's web search budget ran out; the legion-efi and i2c sweep could not run, so those two targets rest on GitHub reads and repository searches only (five repo queries and two issue queries for the window found nothing new). Write-ups: xe, nvgfx, mt7925-wifi-native, mt7925-bt, mt7925-wifi, legion-efi, i2c. Audio, CBMR, CPU power and ITE keyboard had nothing new.

Actionable

  1. macOS 26.7.2 (25G332) breaks MtkWlanNative attach, with logs. zld96969 posted five captures on mt7925-hackintosh-wifi#4: the kext stays out, MtkWlan falls back to its Ethernet port, mtkwifi still joins. The MediaTek Bluetooth USB device is also missing on that build. Needs the abi/native-wifi dump redone for 25G332 and a LegionBtFix check. The README of MtkWlanNative now says 25G332 is unsupported (d802610). Logs filed under logs/zld96969/. High on the symptom, Medium on the cause.
  2. VoodooI2C master has no VoodooI2CPCILakeController (054d5d1), so LegionI2CArrowLake only binds on the 2.9.1 release. README fixed and the ASUS FX608LP repo and bafcz credited (6d6c919, 3c3c9e2). Answered on LegionI2CArrowLake#1. High.
  3. Mountainatic/ArrowWhatever: a second 275HX with the 7D67 iGPU on XeGfx, with ten patches (a userptr pin versus unwire race, a GGTT zero-copy flip deferred until the GPU is idle, about 15 to about 60 fps). Patches are PolyForm Noncommercial, so ideas only; read patch 0001 against I-116. Credited in the XeGfx README, which also stops claiming that one machine has run XeGfx (534220c). Medium.
  4. SimpleMtkBtFw#2: bluetoothd on Tahoe waits for LE PHY and Data Length events that MediaTek controllers never send, so BLE HID pairing stalls (issue). Test BLE HID on our MT7925. Medium.
  5. NullMoth translator patch from eisda (#419, LGPL, 53 failures down to 0) and the 0.6 s frame miss (#410) to compare with our flip traces. Medium.
  6. Arrow Lake EXITBS:START hangs on other machines (1401#110, skybreak, 1401#58, #64, #115): VT-d off, AppleXcpmExtraMsrs, a MADT with too many processors. Record the Legion's BIOS VT-d state against I-118. Medium.

Notable

  • macOS 26.7.2 release notes (Apple) list Bluetooth, Wi-Fi, USB, IOKit and kernel fixes; no public report ties them to the missing Bluetooth device.
  • NullMoth 2.6.2 is the newest release, no 2.6.3; #406 (RTX 5050 Laptop translator error), PR #418 (AIR !dbg parse), #412 (RTX 5080 corner resize still stutters).
  • WhateverXe 0.6.1 and WhateverXe-Graphics 0.3.2: ADL and RPL, experimental eDP, DEBUG stack-usage fix; Xe-LPG still not planned.
  • jwx29/NvGfx (fork of ours, Meteor Lake 155H) keeps pushing; its README mentions a vm_page_validate_no_references panic. In the outreach queue already.
  • mnnobre/AirportMtkWlan#2 (MT7922 kext not loading on 26.7.1, MaxKernel 24.99.99).

New projects

Repo Verdict Relation
Mountainatic/ArrowWhatever related, credited in XeGfx same hardware
Mountainatic/YaoShi16-Ultra-OpenCore-efi related (MIT) same hardware
ydd1126/skybreak related, not credited yet same problem
JoMei9019-real/AirPort-RTW related, credited in MtkWlanNative studied
HungLe17/AirBridgeRTL related same problem
KangweiZhu/ROG-G16-2024-Hackintosh-QEMU-KVM related, marginal studied

Nine other candidates were noise (laptop EFI dumps, a Vulkan AI runtime, a Forza launcher).

Conversations

  • Posted: LegionI2CArrowLake#1 (VoodooI2C master and the polling question) and mt7925-hackintosh-wifi#4 (acknowledged, logs filed). Both logged in watchthelight; the zld96969 thread now waits on them.
  • Follows added for ArrowWhatever, AirPort-RTW, AirBridgeRTL and skybreak.

Logs ingested

Five terminal captures from zld96969 (macOS 26.7.2 / 25G332, MT7925 on an ASUS ROG Strix B860-G board), through the redactor. The first pass left a shell prompt (user and host name) in the text, so the redactor now replaces user@host prompts, with a test (tools/redact.py, tests/test_redact.py), and the files were written again.

Public edits

Held back for the main session: the NvGfx docs/XE.md mirror of the XeGfx edit, and the tracker research rows.

First run (Windows, 2026-10-23 23:28 to 2026-10-27 09:29 UTC)

Already handed to outreach by the first run; kept here unchanged.

Full run, all eleven targets. Windows: 2026-10-23 23:28 UTC to 2026-10-27 09:29 UTC for every target except xe (from 2026-10-24 12:36 UTC). 379 API calls to collect, 0 errors; deep reading split over five readers.

Actionable

  1. macOS 26.7.2 (25G330) and our Wi-Fi and Bluetooth. A user reports both gone after the update (mt7925-hackintosh-wifi#4). MtkWlanNative only loads on builds it has headers for (25G241), and LegionBtFix was patched against 25G241 too. Needs headers dumped on 25G330 and a LegionBtFix check, both on a Mac. See mt7925-wifi-native and mt7925-bt.
  2. Our drag fix measured on an RTX 5080, with an OpenGL regression (nullmoth#412): the stalls go away, but two state-cache commits break Alacritty text. See nvgfx.
  3. zep245/IrisXe-macos-driver: a working i915 and ANV driver on Tiger Lake with written answers for a sandbox problem like our I-106 and kernel panics like our I-116. See xe.
  4. Brightness persistence (PR #312) is verified on an RTX 5080 Laptop; unmerged and unlicensed, so ideas only.

Notable

  • HDR on NullMoth is blocked by CoreDisplay's force8BPC for class names containing "NVDA" (#341).
  • HiDPI: the missing kDisplayModeValidForHiResFlag and the 12-byte override format confirmed again (#354, #341, #360).
  • The shader-cache write cost comes back after every driver update (#370).
  • NullMoth 2.5.5 to 2.6.2 (2.6.x is installer-only); mickys' translator branches; a 0.6 s frame-miss report (#410).
  • WhateverXe 0.6.0 (Xe-LPG and Xe2 not planned), WhateverXe-Graphics 0.3.0 (AIR conversion), JulioNectar power, leak and backlight work, metal2vulkan's coverage table and Arrow Lake ANV profile, drm-kmod's Arrow Lake-S fix.
  • LenovoLegionLinux Gen 10 EC offsets (tested on 83LY), VoodooI2C's first code change in eleven months, the AW88399 amp work for the Y9000P 83F5 (probably not our hardware).

New projects

Repo Verdict Relation
zep245/IrisXe-macos-driver related, credited in XeGfx studied
CENSEDAXIS20025/ASUS-FX608LP-Hackintosh-Multiboot related, credited in XeGfx same hardware
dssoftx/Hackintosh-X1Carbon-G11-IrisXe-Tahoe related studied
HackSucks/mtwlm related, credited in MtkWlan studied
thanhnndev/mt7925-macos-wifi related, credited in MtkWlanNative same problem
DGBooth/AirportRTW89-MT7922 related, credited in MtkWlanNative studied
Sunburst0218/MECHREVO-GM6IX7B-i7-14650HX-RTX4070-Hackintosh related same problem
pranish-dev/loq_hackintosh related same problem

32 other candidates were noise (laptop EFI repos, unrelated apps).

Conversations

Logs ingested

None. The only labs issue in the window (#4) has no attachments yet.

Public edits

  • XeGfx credits: be1744e
  • NvGfx xe docs/XE.md mirror: 687f8d5 (the gpu/xegfx subtree pull was skipped: NvGfx xe carries XeGfx 0.0.44, which XeGfx main does not have yet)
  • MtkWlan credits: 4316175
  • MtkWlanNative credits: 8895e4d
  • Write-ups: b27e44d