xe: 2026-10-30 to 2026-11-11

Read from GitHub only. Nothing below is reproduced on our machine unless it says so.

Mountainatic/ArrowWhatever (Core Ultra 9 275HX, 7D67, the same chip as ours) added 28 patches on 2026-11-02, 11-05 and 11-06. We had ported 0008 and 0009 to a branch. The new ones cover four things we have open:

• Hardware cursor and scaled sizes. 0023 and 0025 let i915 take over the eDP pipe, write a cursor plane directly and use the plane scaler for HiDPI sizes. The joiner and 240 Hz parts are for that panel, not a 2560x1600 one.
• Backlight. The same patches call i915's intel_backlight_set_acpi() and add an IODisplayParameterHandler for the slider. It needs i915 to have found a backlight on the panel.
• Colour stages. 0030 reports AGDC colour stages and a BGRA plane; without it CoreDisplay runs one more full-screen copy per frame.
• GPU clock floor. 0035 and 0036 raise the SLPC minimum frequency when frames come late. The author's numbers: 126 fps with it, 52 to 68 without.

Also useful: a per-CPU MMIO watchdog (0028, 0029) that names the register read that never finished.

Not in there: anything for the GuC GGTT top, sleep and wake, or chips other than Arrow Lake.

The pitch mismatch on 3120-wide panels (WTLIntelGfx #14) is not fixed. 0024 reads PLANE_STRIDE and refuses the flip when it differs, so those frames take the copy path. The thread is ArrowWhatever #2.

Confidence: High that the patches contain this (diffs read), Medium that they work as described (one machine, the author's own numbers). The patches are PolyForm Noncommercial, like the tree they apply to.

• hepengfei00/XeMac: 7D67 again, with an external DisplayPort screen at 4K60 HiDPI. It pins our xe branch plus ArrowWhatever 0001 to 0037 and adds 0038 to 0046: DisplayPort and AUX mapping, WindowServer framebuffer fetch, native timing published to AGDC. It credits us and keeps the notices. Screenshots and docs, no raw logs. Medium.
• taqieallah/ThinkPad-E16-ArrowLake-macOS: 7D51, a diff on WTLIntelGfx 0.0.47 with a zero-copy flip, 43 to 53 fps reported. Open problem there: the CPU sits near 550 MHz. Medium.
• change-48/XEGFX-MIGRATE: the bring-up log behind #14. Findings: a reserve_bootfb of -28 means the GGTT allocator does not know the boot framebuffer is in use and the accelerator overwrites it; plane geometry registers only latch when PLANE_SURF is written; a frame at p50 24.8 ms splits into 10.84 ms in the driver and 14.10 ms after the signal. Medium.

zep245/IrisXe-macos-driver (Tiger Lake 9A49, releases 1.1.4 and 1.2.5) has an accelerated WindowServer confirmed by two outside testers in its issue 1. Its DESIGN.md records how scaled sizes were made to appear (commit c71dffc): publish IOFBScalerInfo and IOFBTimingRange, accept programmed timings, allocate the scan-out buffer larger than the panel and let a plane scaler shrink it, set kDisplayModeValidForHiResFlag, and put the panel EDID on the display services. Override plists alone do not add sizes. The same file has notes on the brightness parameters and on handing power-state calls to IOFramebuffer for wake. Medium, one laptop. Video decode and power numbers there are the author's own.

• Two hybrid gaming laptops only got a working slider through the firmware's WMI brightness method (\_SB.PC00.AWMI.WMAA), not through the panel: WhateverXe-RaptorLakeP and NootedGreen 0326434. Whether a Legion 5 15IAX10 has that method is not checked. Low for our machine.
• WhateverXe #70: a panel resets DPCD 0x720 and 0x721 after display doze, so AUX brightness has to be re-asserted.
• WilliamAlexanda 43f626e: brightness capped at 18.5% because the driver's PWM period differed from the firmware's. Read the period before writing a duty.

For WTLIntelGfx #4:

• #315 is still open, quiet since June.
• #418 is merged: GMADR init on Meteor Lake and newer, media GT1 skipped, HDCP off.
• #468 was replaced by a scatterlist fix in freebsd-src (a6b40d7) and by #534 (open) for the fictitious BAR range.
• Three LinuxKPI fixes bear on hangs: 79e290d (cancel_delayed_work_sync, a Meteor Lake tester says it ended a sporadic hang), 2ff9182 and f739d9f.

High for the state of the pull requests, Medium for the effect.

• WhateverXe 0.8.0 (2026-11-07): nothing for Meteor or Arrow Lake.
• WhateverXe #78: CHICKEN_MISC_2 bit 30 has to be set or tiled planes show black. Worth a look after resume.
• WhateverXe #85: a Legion Y9000P with a lit backlight and no image until one lid cycle; the only difference seen is DPCD 0x0205.
• sulthonuladib (Alder Lake 46A6) got stages 1 to 3 with adlp_guc_70.bin linked in by hand, then moved to a patched Apple driver.
• pongpan-bk's Core 5 120U driver has committed object files, obfuscated constants and no licence. We did not build or run it.
• Web search for the window found nothing new.