[sithglan]

Base: integration/aurora (050f6a5). One commit.

The 27.2 AOP firmware (AppleAOPAudioFirmware-610.18, K116s) advertises AFK endpoints 0x20-0x27 and 0x29-0x30. Its boot stage setup_endpoints brings them up one at a time and waits for the host's STARTEP on each. The driver started only 0x20-0x2e, so the firmware waited for 0x2f forever ("waiting for endpoint enable, _routing=0x2f" in its log). It never reached system_ready/services_ready and announced no EPIC service at all: no aop-audio, so no sound card, and no sensors.

AFK_ENDPOINT_COUNT goes from 0xf to 0x11. start_afk() still skips an endpoint the firmware does not advertise (has_endpoint()), so older firmware with fewer endpoints is unaffected. Every T6050 machine that moves to the 27.2 stub (J714s and others) needs this.

Tested by Jens on the K116s (27.2, 26B8125): the firmware announces its 20 services and the sound card registers.

Build: arm64, LLVM 22, Rust; drivers/soc/apple/ with Jens's K116s config, no new warnings.

Authorship. Jens Axboe wrote these commits and runs them on his K116s (16-inch MacBook Pro, M5 Pro, OLED; Mac17,12; stub macOS 27.2 26B8125). They come from his K116s bundle of 2026-11-11 (integration/aurora 254c9bd plus 34 commits, kernel 7.3.0 running on the machine). I rebased them onto integration/aurora 050f6a5 and removed the LLM co-author and session trailers. Jens has given his OK: the commits carry his Signed-off-by.

[RyanTheTide]

Endpoint start, stop and sleep all key off the count, and has_endpoint() keeps older firmware on the endpoints it advertises. The masks are u32, so 17 endpoints fit. Approving after a run on our M1/M2/M4/A18 machines.

[RyanTheTide]

Interim results. Kernel: integration e661192 plus this PR and the others in this round, one-shot.

• M4 Pro (J616s) and M1 Pro: AOP comes up ("AOP IS READY"), no AOP/AFK warnings, and HPAI microphone capture works before and after 3 s2idle cycles.
• 0 AVC, 0 kernel errors.

The M2 and A18 Pro runs are still queued.

[RyanTheTide]

M2 results (T8112 MacBook Air, j413), kernel integration e661192 plus this PR and the others in this round, one-shot, against the production kernel on the same base: AOP comes up ("AOP IS READY") with no AOP/AFK warnings, HPAI microphone capture works before and after 3 s2idle cycles, and AOP stays clean after each wake. 0 AVC, 0 kernel errors. With the M4 Pro and M1 Pro runs posted earlier, the A18 Pro run is the one still to come.

[DjDeveloperr]

Approving. start_afk() already skips endpoints the firmware's RTKit map doesn't advertise, so raising the ceiling to 0x30 only adds 0x2f/0x30 where a firmware lists them. The per-endpoint tables are sized by the constant, the u32 masks cover 17 entries, and the explicit T8140 endpoint lists are unchanged. It merges cleanly on current integration/aurora; I updated the branch so CI runs against the current base. The hardware evidence in the thread covers this exact change: K116s announces its services and registers the sound card, and M1 Pro, M2 and M4 Pro show no AOP regressions. This also fixes the T6050 stall @joshuaswarren reported in #1850.

Follow-up idea, not for this PR: warn when the map advertises an app endpoint above the table, so the next firmware that grows its map fails loudly instead of stalling.