watchthelight labs

mt7925-wifi: 2026-11-11 to 2026-11-13

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

  • AirportMtkWlan shipped five releases in two days (v2.1.0 to v2.1.4), all for MT7921, MT7921K and MT7922, not MT7925. It is the closest relative of our planned MT7921/MT7922 backend: an OpenBSD mwx(4) port wrapped in a native Tahoe Wi-Fi front end. The fixes it lists are a useful list of pitfalls for that backend: 2x2 stream capabilities left at 1x1 after firmware load, an edge-triggered MSI interrupt that can strand until a 1 second watchdog, background scans going off-channel during transfers, EDCA/WMM never programmed, and PCIe ASPM and L1 substates left on (v2.1.4 clears them on the card and its upstream bridge). A tester on MT7921 still reports under 40 Mbps after v2.1.4, so these were not the whole story. Licence is GPL-3.0, so this is ideas only for our ISC drivers. We already turn ASPM off in MtkFw.cpp (aspmOff), so that item is not new. Confidence: High that the releases say this, Medium that it carries over to our backend. Links: https://github.com/Arxusion/AirportMtkWlan/releases/tag/v2.1.3 , https://github.com/Arxusion/AirportMtkWlan/releases/tag/v2.1.4 , Arxusion/AirportMtkWlan#2

  • v2.1.1 fixes the same Tahoe "missed beacons" leave every 60 seconds that we fixed in the native front end. Their cause: the link quality beacon update (the 0x1dc byte snapshot) left its "snapshot valid" and "counters valid" flags at 0, so the Family code ignored it and the association timer fired after about 60 seconds. They now fill RSSI, per-core RSSI, SNR, noise and the beacon counters, and return a -95 dBm noise floor to stop APPLE80211_IOC_NOISE errors. We reached the same fix independently (a 0x27 update of 0x1dc bytes every second, in 0.4.9). This is corroboration, and a field list we can compare against ours in case our snapshot leaves noise or SNR at zero. Confidence: High that they changed this, Medium that our snapshot has any gap. Link: https://github.com/Arxusion/AirportMtkWlan/releases/tag/v2.1.1

  • Issue #11 on AirportMtkWlan is the same failure class as our warm-restart loss (I-129, WTLWireless #1). An MT7922 on Sequoia freezes a few seconds after the radio comes up. On every later boot the driver logs driver own failed with lpctl=0x00000004, and neither a WFSYS reset nor a PCIe function-level reset clears it. Only removing AC power does. Our MT7925 stops at W3 fail: patch download after a warm restart and needs a cold power-off, and our 0.11.4 plan is to take driver ownership and reset WFSYS from a shutdown hook, as Linux does. This suggests the stuck state is family wide and that in-driver recovery after the fact may not be enough, so the shutdown quiesce matters more than a retry loop. It is a different chip and a different trigger (a hard freeze rather than a clean restart), so Medium. Link: Arxusion/AirportMtkWlan#11

  • A third party documented MT7921 Bluetooth 0489:E0CD on Tahoe by patching the supported-device table inside bluetoothd. The repo says Tahoe 26.x bluetoothd holds a hard-coded list of seven MediaTek USB IDs (five Foxconn 0489 IDs, 0E8D:7922, 13D3:3568), and a Lilu plugin swaps one entry for the new ID with an equal-length byte patch, with the firmware loaded by a kext. That is the same family of approach as our LegionBtFix (patching bluetoothd bytes on the fly), and it confirms the table exists. Our device is 0489:E111, which we already handle. GPL-3.0 (a fork of laobamac/SimpleMtkBtFw), zero stars, two commits on 2026-11-13, so Unverified beyond the README's own claim. Link: https://github.com/sl520987456/MT7921-Bluetooth-Tahoe