xe: 2026-10-27 (full run, window 2026-10-24 12:36 to 2026-10-27 09:30 UTC)

Nothing in this window supports Arrow Lake or any Xe-LPG part on macOS.

zep245/IrisXe-macos-driver: a working i915 and Mesa ANV driver for Iris Xe. The repo is a fork of NullMoth's Intel path that runs on Tiger Lake 9A49 under Sequoia 15.8.1: FreeBSD drm-kmod i915 and LinuxKPI inside the kernel (NMXGPU, NMXAccel, NMDRM), and a Metal plugin on Mesa ANV with a translator descended from metal2vulkan. Display, cursor, brightness, PSR1, RC6 and sleep work; idle power is 5.2 W. Tahoe is planned. That is the same shape as XeGfx, one generation older. Two parts of its notes line up with our open problems:

• Sandboxed apps get no Vulkan device because the sandbox refuses opendir(/dev/dri). They answer it with a hook in the plugin and a shader cache the sandbox can write to. That is our I-106 from the other side.
• Its design notes list the in-kernel crashes it hit and their causes: the thread_call group thread limit, a preemption_level -1 panic from cond_resched inside ggtt_pin and shmem_get_pages, a TLB flush timeout and a machine check at the stolen memory base. Our I-116 died under heavy I/O; these are the first places to compare.

Licence is PolyForm Noncommercial 1.0.0, so ideas only. High confidence on what it is, Medium on the mapping to our issues.

WhateverXe 0.6.0 (release): Alder Lake and Raptor Lake H, P and U, fixed HDMI routes from the VBT, experimental fixed Type-C DisplayPort, and a commit that defers Display 13 framebuffer opens until AGDC registration (e98d3f7). The README now lists Xe-LPG (Meteor Lake) and Xe2 as not planned. The AGDC ordering is worth reading next to our AGDC readback work. High confidence.

WhateverXe-Graphics 0.3.0 (release): native AIR conversion through the system compiler fixes shader compile failures from Sonoma to Tahoe. A possible angle on our I-117 (Geekbench translate failure), not checked. No licence file, ideas only.

JulioNectar/Hackintosh-XPS13Plus-IrisXe-Tahoe: render-engine workarounds after every engine start (875ce5d); hybrid P-core power rerouted so HWP units are right, with Geekbench single-core going from 1269 to 1697 and idle package power from 4.6 W to 3.3 W (0492b29); a 96 KB leak per command buffer that grew WindowServer past 10 GB (0698648); eDP backlight over DPCD AUX (453a7b5). Their issue #17 (an A7A8 laptop where the accelerator probes but the pipe never scans out) is a parallel to our 7D67 display pipe. Ideas only under their licence.

steelbrain/metal2vulkan: three published crates, a coverage table (958 of 2135 macOS 27.0 SDK items), docs/REFUSALS.md, and an Arrow Lake ANV validation profile (intel-arl-anv.json). Commits in the window hold shared-event waits instead of handing them to a queue and read YCbCr planes in place on integrated GPUs. The refusals list is where to look for I-117. GPL-2.0-or-later or LGPL-3.0-or-later.

lutzbichler/drm-kmod (pr/7.3): Linux 7.3 stable fixes, mostly amdgpu. For Intel: is_arrowlake_s_by_host_bridge() always returned false (ade98e6), DMC load-address and payload checks, an RPS interrupt unmask race, and a cap on xe_mmio_wait32 delays. No LinuxKPI changes.

CENSEDAXIS20025/ASUS-FX608LP-Hackintosh-Multiboot (repo): the same Core Ultra 9 275HX, 7D67 iGPU and RTX 5070 Laptop as ours. It runs macOS 15.7.8 on the RTX through NullMoth with the iGPU blocked and the BIOS MUX set to discrete only. The opposite of what XeGfx needs, but a useful same-hardware datapoint.

samranahmad09/WhateverXe-RaptorLakeP: a Raptor Lake-P bring-up fork with an i915 register capture next to a macOS readout and a passing blitter ring self-test. A method worth copying, not a driver.

dssoftx/Hackintosh-X1Carbon-G11-IrisXe-Tahoe: Raptor Lake-U on Tahoe 26.7.1 through the JulioNectar and NootedGreen path. Metal and sleep work.

pongpan-bk's 120U driver (no licence, abandoned by its author, claims unverified), kitsuneislife/nmx-intel#1, kumibrr/oc-tahoe-pavilionplus14-ew0000 (now private), steelbrain/reims-vgpu for this target. Discovery: debianita22/AzulFix, peiikai/LASRecomp, and nine laptop EFI repos.

Mountainatic/ArrowWhatever: another 275HX with the 7D67 iGPU, running on XeGfx. The repo was created on 2026-10-25 (first named intel-xe-lpg-igpu-macos, then ArrowBridge, ArrowWhatever since the 27th). The owner has a Core Ultra 9 275HX with an RTX 5080 Laptop on macOS 26.7.1 and ships ten patches on our 5a19a73 (XeGfx 0.0.44), plus a port of the 0.0.35 plugin. The parts that touch our open work:

• A zero-copy GGTT scanout path with a flip deferred until the GPU is idle (their "K9" doc). They report about 15 fps rising to about 60 fps at 2560x1600. That is the same slow-scanout problem we have.
• A fix for a page-lifetime race in userptr buffers (their patch 0001): the panic text is xegfx: vm: freeing wired page ... (ref 0x1) from i915_gem_userptr_put_pages, and a 40 minute soak ran clean afterwards. Our I-116 panic in N-289 sat in a TLB shootdown from the i915 free worker. The two may be one bug or two; nothing links them yet.
• A read-only host pointer check (nvmtl_range_writable) that came from our own tree.

The patches are PolyForm Noncommercial and the scripts and docs are MIT, so we read the patches as ideas only. There are no raw logs in the repo, so the numbers are the owner's claim. Medium confidence. We credited the repo in the XeGfx README and corrected the line that said only one machine had run XeGfx (534220c).

• Mountainatic/YaoShi16-Ultra-OpenCore-efi (MIT) is the EFI of the same machine. It boots NvGfx 2.5.3 boot-args with no Xe kext, and its README says the panel runs through the NVIDIA path.
• jwx29/NvGfx, a fork of our repo (branch meteor-lake-local): a Meteor Lake 155H (7D55) on Sequoia 15.8.1, with AGDC, scanout and VRAM reporting commits, the newest at 18:21 UTC. Its README mentions a vm_page_validate_no_references panic after accelerated WindowServer use. It does not match the rendezvous panic of I-116. Already queued for outreach (thread t169).
• WhateverXe 0.6.0 and 0.6.1 and WhateverXe-Graphics 0.3.2 (b00t0x): ADL and RPL support, experimental eDP and PWM, a fix for DEBUG boot freezes caused by kernel stack use. Xe-LPG and Xe2 are still not planned. Nothing for us unless a stack-depth problem shows up in our own DEBUG builds.
• CENSEDAXIS20025/ASUS-FX608LP-Hackintosh-Multiboot added ten commits, mostly README edits (1401 EFI route, touchpad row). See i2c.

NullMoth 2.5.4 to 2.6.2 changes for video decode only. danielsaluceste/Samsung-NP550XDA-macOS-Tahoe-OpenCore- (Tiger Lake EFI on WhateverXe), Cognitus-Lab/AximOS (a cross-vendor Vulkan runtime, no macOS GPU driver), two JulioNectar issues, WhateverXe forks. No commits in the window for freebsd/drm-kmod, zep245, VedCloud, LunarBridge or nmx-intel, and no outside issues on XeGfx.