[RyanTheTide]

fleet11 (aurora-wip 8348f82) had two regressions against fleet10 on real hardware. This branch is the exact source of the kernel now on the Neo and the J616s (7.3.0-aurora-fleet11b, ab3a90c). Each commit holds back one merged change until the thing it depends on has shipped. Draft because each one is a decision for Ryan, not a fix to merge as is.

1. SEP biometric ABI v5 (f70f1ac and follow-ups)

The series moves SEP_BIO_IFACE_VERSION from 4 to 5 and grows sep_bio_enrol_start from 136 to 144 bytes. Every packaged userspace still speaks v4 and compares the version for equality: aurora-sep-sensor-present (fedora 1-14, alarm-aurora, nixos-aurora) and the libfprint SEP patch. On the Neo with plain fleet11, GET_INFO returned version 5, so aurora-sep-ready and fprintd skipped themselves (ExecCondition failed) and Touch ID was gone. The kernel side was fine ("restored 3 enrolled identities").

The v5 libfprint is still the draft aurora-silicon/aurora-sep-userspace#3, and the convergence report already listed this series as not ready for the same reason. The commit puts the SEP driver files and the UAPI header back to what integration/aurora ships (v4). Drop it once the v5 libfprint and a sensor check that accepts 5 are packaged on all three distros.

2. #628 driver commit on the J616s (3db9b1f)

#628 says the J616s Stage 2 DTB must carry its DT commit (dcpext1 and its DARTs and mailbox get power-domains = <&ps_dispext1_fe>) before a kernel with its driver commit boots there. The J616s still runs the old Stage 2 DTB. A fleet11 kernel with the driver commit did not come back from two one-shot boots: both times GRUB fell back to fleet10 about 25 s later than a normal reboot, which fits an early panic plus panic=10. The same image without the driver commit booted at once. The revert keeps the old behaviour (the -22 for power-controller@490 is still logged) until the J616s Stage 2 DTB is refreshed.

Testing (both machines, full dmesg in reports/fleet11-20261031/)

• Buildbox: run-host-tests 152/152, apple-dcp selftests 80/80, asahi-m4 selftests 23/23.
• Neo, one-shot with the old options, one-shot without clk_ignore_unused pd_ignore_unused idle=wfi arm64.nowfxt, then default: AVE H.264 and HEVC 1080p 300/300 frames through ffmpeg and GStreamer, v4l2-compliance 48/48, no -p 3 lines, no sync_state lines, Wi-Fi and BT through the DT bootstrap, ISP camera 60 frames, Touch ID with 3 enrolled fingers, admac and i2c runtime-suspended after idle with audio and the MTP keyboard still there, 3 of 3 rtcwake cycles, aurora-diag check against the fleet10 snapshot passes.
• J616s, one-shot then default: GPU, Wi-Fi, three displays, Touch ID, audio muted with the HiFi profile and speakerguardd attached, the backend-DAIs line gone (ASoC: apple: macaudio: no PCM for the J616s route-only front-ends #591). Only the two power-controller@490 lines remain, as expected with the old DTB.

🤖 Generated with Claude Code

[RyanTheTide]

Superseded by the fleet12 carry (https://github.com/aurora-silicon/linux/tree/fleet12/carry-20261031). Touch ID v5 shipped, so the SEP v4 hold is gone, and the #628 driver revert is carried per fleet build until the Stage 2 DTBs are refreshed. This hold branch was never meant to merge.