MediaTek MT7925 Wi-Fi: what moved, 2026-10-22 to 2026-10-23

Game-5040/hackintosh-hp-8f06-macos26 (MIT, in Chinese) is an HP OMEN 16-ap1xxx: Ryzen 9 9955HX, RTX 5070 Laptop, MT7925 Wi-Fi 14C3:7925, Bluetooth 0E8D:8C38, macOS 26.7.1 (25G241), OpenCore 1.0.8 from NullMoth's 1401 app. It runs MtkWlan 0.11.3, MtkWlanNative 0.4.11, MtkBt 0.2.4, BlueToolFixup 2.7.2 and LegionBtFix 0.2.0 with mtkfe=native mtknativeif=1, kext order and MinKernel as in our READMEs. The firmware has no S3, so nothing there covers sleep. Seventeen commits on 2026-10-23 added a pitfall table and a version section; the parts that are feedback on our drivers:

• Saved network stops joining (14581e5c, README line 634): a previously joined network spins and fails; Forget and rejoin fixes it, sometimes with Private Wi-Fi Address switched off for that network. New report, no logs; the private-address lever points at the set_mac path. Medium.
• Slow scans and intermittent DFS networks (same commit, lines 637 and 813): they read our XZ country code correctly. setCOUNTRY_CODE is recorded only (src/MtkWlanNative.cpp 439-455) and the core always runs the US domain with 52-144 passive at a 120 ms dwell (src/MtkScan.cpp 489-546), so an AP on a DFS channel is caught at most once per sweep. Their workaround, pin the router to 36-48 or 149-161, is right for a US domain. The XZ fact was in our README; the scan effect was not. High on the code, Medium on the effect.
• Instant Hotspot entry does nothing (c8f770bd, line 636): the Continuity hotspot entry in the Wi-Fi menu does not react; joining the same SSID through Other... or networksetup with the password works (iPhone on Maximize Compatibility, 2.4 GHz WPA2). They blame AWDL; discovery of that entry is BLE, so the cause is unverified (itlwm had the same symptom, #491). If that manual join was WPA2-PSK, it is the first on-air WPA2 join through the native front end; we asked them to confirm.
• Bluetooth (14f8e5c1, lines 635 and 764-815): a third tested id, 0E8D:8C38, on 0.2.4: MtkBtState ready, 3 sections, 447,872 bytes, On as THIRD_PARTY_DONGLE, with LegionBtFix on a non-Lenovo machine. On 0.2.2 Bluetooth switched itself off and on, with bluetoothd crash reports showing SIGBUS in TxLoop. Their attribution ("the 0.2.2 reset loop") is off: the reset loop was 0.2.3's and 0.2.4 fixed it; 0.2.2's bug was the warm-restart half-load. The SIGBUS site is new and not the LE-connect crash LegionBtFix patches; the crash report is not in the repo. Symptom High, cause Unverified.
• AirDrop feasibility (docs/airdrop-feasibility.md): desk research that quotes our native #2. Useful reference list: Owly (MIT, sans-IO AWDL state machine, MT7925 is its first supported card on Linux), mt76-usb-macos (Clear BSD, user-space MT7921U/MT7925U over libusb with experimental injection), OWL (GPL-3.0). Their effort estimates are guesses; their affected-feature table overstates (Handoff and Watch unlock are BLE). Low beyond the references.

Not about our kexts but worth knowing: NullMoth 2.5.2 keeps no brightness across boots and the panel jumps to maximum a few hundred ms after a display wake; their MIT tools/brightness watchdog holds it down, and upstream PR #312 is the real fix. 1401's USB map writer keys ports by class name, so a second port on the same controller overwrites the first (README section 15); no upstream issue exists for that.

• Apple's own MediaTek driver. macOS 26 ships com.apple.AppleSunriseWLAN.dext (ipsw-diffs, 26.6.2 to 26.7). On tonymacx86 (thread 332855) a kext port of it runs three MT7922 cards on 25G241 with SIP on (two users); MT7925 is "planned". What the dext matches and whether it can drive an MT7925 are unverified; the port is a binary built from Apple's code, so nothing there is usable as code.
• eiyuumiru/SimpleMtkWlan v0.3.0-beta (GPL-2.0, ideas only), RZ616 on 15.8.1 and 26.7.1: beacon-loss events (0x13) were discarded, so after the AP went away the interface stayed joined with the link up (a69355d2); deep sleep was switched on before RUN and joins to fast APs timed out (81e4232); cacheable bounce buffers took throughput from 41 to 96 Mb/s (9eff5442). The first is a candidate cause to check for our traffic stall after about 38 minutes. Medium.
• MT7922 and RZ616 options for testers (#2): YYavicWiFi (14c3:7922/7961/0608, native menu on 15.8.1 only, author-measured 913 Mb/s), AirportMtkWlan-RZ616 (14c3:0616 on Tahoe, source only, unproven), and three SimpleMtkWlan forks adding 14c3:0616 in two days. None uses our code.
• Firmware: linux-firmware updated the MT7925 Wi-Fi pair on 2026-10-20 (5a3f625f). Not tried.

• New project: Royiii3/r9000p-adr10-hackintosh-notes, a Legion R9000P ADR10 (Ryzen 9 8945HX, MT7925 14C3:7925, RTX 5060 Laptop). macOS 26 reached the desktop on NullMoth 2.3.4 only with cpus=1; Wi-Fi, Bluetooth, audio and backlight never worked, and the author removed macOS. No Wi-Fi attempt is recorded, so there is nothing to learn about the card. High.