watchthelight labs

nvgfx: 2026-11-11 to 2026-11-13

Read from GitHub only. Nothing below is reproduced on our machine.

  • No new upstream code since 2.9. The newest commit on NullMoth's main branch is dated 2026-11-10 (changelog for 2.9), and no 2.9.x release exists. The stable release is still v2.9. Nothing to rebase onto yet. Confidence: High. See commit 3d8ffab and the release list.
  • v3.0.0-beta.1 and Liberator (published 2026-11-11, just before this window). A pre-release that ships a Mac app, Liberator, to move an existing NullMoth setup from macOS 15 or 26 to macOS 27.2. The release notes say it was tested on one desktop with an RTX 50 card and that laptops are untried. It disables AMFI and lowers SIP for macOS 27. Not relevant to a macOS 15/26 build for now. One laptop report already exists: #563, an early "Non-monotonic time" panic on a 12th-gen Intel laptop, where the reporter's CpuTscSync was skipped on macOS 27 by its MaxKernel limit (the reporter has not tested that as the cause). Confidence: High for the release facts, Low for the cause. Release.
  • Hybrid laptop mode (nmhybrid) is unchanged. #555 (boot panic in NMIntelFB on Alder Lake-H, then head 4 pipe inactive) is still open with no comments or maintainer reply after 2026-11-10. No commit touches it. Confidence: High.
  • Laptop brightness on the NVIDIA panel is a family problem. #551 (82JU/82JW Legion, NBCI GETBACKLIGHT empty, no DPCD brightness) got a second data point on 2026-11-12 from a Legion Y9000P 2022 (82RF, RTX 3060 Laptop, dGPU-only mode, driver 2.8.1). There, NBCI _DSM functions 0 and 0x10 return data but function 0x14 returns 0 bytes, while the static ACPI _DSM function 20 returns a full package. That points at the firmware call path rather than the driver's opt-in or GUID match. Why it matters: our panel is sometimes driven by the NVIDIA GPU, and our logs show a slow gap after the NBCI _DSM call. Confidence: Medium (reporter's logs, not reproduced).
  • Backlight not restored on restart, and HDMI hotplug, on a MUX laptop with 2.9. #582 (NVRAM variable reads 02 ff ff after restart; the reporter runs DisableVariableWrite with no emulated NVRAM, so the cause may be the OpenCore setup) and #581 (an HDMI display plugged in after boot never gets a connect interrupt; the log says the head gets 1360x768 and nothing follows). Both on an Acer Nitro 16 with an RTX 4060 Laptop. Weak evidence, useful only as a hybrid-laptop symptom list. Confidence: Low.
  • WindowServer spends CPU in surface share/import ioctls during video playback. #566 (the duplicate #565 was closed as not planned) has a spindump: nvmtl_vk_surface_mark_newest to nvRmApiSurfShare, plus dma-buf imports, with os_map_kernel_space log lines rising from hundreds to thousands per minute. The reporter tried NVMTL_SURF_IMPORT_CACHE=1 but changed other settings at the same time, so the effect is unproven. This is a desktop RTX 4070 Ti on 2.9. It may share a cost with our window-drag stall issue (#11), but our trigger is the first minutes after login, not video. Confidence: Low for any link.
  • PAT patch swap fixes GSP init on a Legion laptop, second confirmation. In #69 a Legion R9000P 2021 (RTX 3070 Laptop) on 2.9 and macOS 26.7.3 gets past an rm_init_adapter failure by switching the OpenCore patch from Shaneee/Zormeister to Algrey/Zormeister _mtrr_update_action and cold booting; the same user posts a works report. Not our hardware generation (Ampere, AMD CPU). Confidence: Medium.
  • HiDPI entries: our byte-order finding confirmed on Sequoia. On #455 a Redmi G Pro user reports the big-endian 12-byte DisplayProductID entries make scaled modes appear on macOS 15.8.1, and we have asked two follow-up questions there. On #354 we asked someone on 2.9 for a CGDisplayCopyAllDisplayModes dump. Already known; both threads wait on the reporters. Confidence: Medium.
  • New public EFIs from NullMoth users on hybrid laptops. HP Victus 16 (RTX 4060 Laptop, i7-13620H) reports Metal 3 with the internal panel plus an external display, brightness working with nvrmacpibl=1, and sleep not resuming; its README says the Intel iGPU is much slower. Legion Y9000X IAH7 runs 2.9 and logs that repeating a power-source SET every three minutes on 2.8.1 caused display flicker and tearing, so its agent now reads the state first (commit deccd17). Both are self-reports from the owners' own machines. Confidence: Low.
  • For the Intel side (separate scout report): the Dell G15 5530 repo's WhateverXe commit 9609e16 adds render forcewake while the render engine executes, as a candidate fix for a UHD hang, and an optional eDP PSR1 setting. Different hardware generation from ours; noted only because we track an Intel hang of our own. Confidence: Low.