Skip to content

M3 Air (J613/J615) Touch ID: SEP and Mesa values from the ADT, and a test machine #79

Open

Description

@scottjones

Drafted with an AI assistant (Claude Code); I reviewed it before posting.

@iconidentify: is anyone working on Touch ID for the M3 (T8122)? If not, we'd like to help. Our 13" M3 Air (J613) is available for testing, and below are the values a T8122 profile and device tree would need.

Today on 2026.10.11.13, the driver loads and stops: apple_sep: no platform profile for this SoC; the SEP stays disabled. t8122.dtsi has no spi2, SEP, SEP DART or Mesa nodes; pinctrl_nub, ps_spi2 and ps_sep are already there.

The values come from Firmware/all_flash/DeviceTree.j613ap.im4p and DeviceTree.j615ap.im4p in the public UniversalMac_26.6.2_25G83_Restore.ipsw. As a check, the same parser run on that IPSW's DeviceTree.j415ap.im4p gives exactly the J415 values in aurora-silicon/linux#152. J613 and J615 are wired identically: only phandle numbers differ.

J415 (T8112, works) J613 / J615 (T8122)
spi2 0x235108000, IRQ 750 0x2a1108000 (ADT 0x91108000), IRQ 767, spi-1,spimc
Mesa biosensor,mesa, R3, 8 MHz, CS 0 same: R3, spi-frequency 8 MHz, CS 0, same reg bytes
Mesa power AP GPIO 178 (/arm-io/gpio) NUB GPIO 29 (function-mesa_pwr → /arm-io/nub-gpio, i.e. pinctrl_nub), active high
Mesa data-ready AP GPIO 177, flags 3 AP GPIO 199 (/arm-io/gpio0), flags 3
Mesa extras none function-monitor-current-3v0 (SMC PMU key), monitor-power-interval = 1000
SEP iop-sep,ascwrap-v4, 0x25e400000/0x6c000, 0x25e050000/0x60000 iop-sep,ascwrap-v6, 0x29e400000/0x6c000, 0x29e050000/0x60000, IRQs 318–321, no power gates
SEP DART dart,t8110, vm-base 0x4000, vm-size 0xffff0000 dart,t8110 at 0x29cac0000, IRQ 324, vm-base 0x100_0000_4000 (the BIT(40) alias, as on J700), vm-size 0xffff0000
/defaults cpx-encryption-mode = 2 cpx-encryption-mode = 2

(Addresses add the /arm-io offset 0x210000000, which matches pinctrl_nub at 0x2e41f0000 and spi1 at 0x2a1104000 in t8122.dtsi.)

So the T8122 profile would most likely be T8112's warm-registration profile with T8140's wide_dma_mask. The two things I can't tell from the ADT:

  • whether ascwrap-v6 changes the SEP mailbox;
  • whether iBoot leaves the sensor powered when the power line is on the NUB controller.

This Air's firmware: system 27.0 (mBoot-20457.1.29), OS 26.6.2 (mBoot-18000.161.10).

What we can do:

Activity

  1. scottjones commented on Oct 12, 2026

    @scottjones
    CollaboratorAuthor

    Drafted with an AI assistant (Claude Code); I reviewed it before posting.

    @iconidentify: Touch ID works on the 13" M3 Air (J613). On 2026.10.11.15 with a T8122 profile and device-tree nodes built from the values above, the SEP attaches, the sensor comes up, and a finger enrols, matches and still matches after a reboot. macOS Touch ID still works afterwards. Patch: scottjones/aurora-linux@t8122-touchid (a51a1e02), one commit on top of 6c982dda.

    Patch:

    • profile.rs: a T8122/J613 profile, which is T8112 with wide_dma_mask: true. Only apple,j613 and apple,j615 select it, because the other T8122 boards' sensor wiring hasn't been checked.
    • t8122.dtsi: sep_dart (0x29cac0000, IRQ 324, apple,dma-range = /bits/ 64 <0x10000004000 0xffff0000> as on t8140), sep, sep_mbox (apple,t8122-asc-mailbox, IRQs 318–321), and spi2 (0x2a1108000, IRQ 767, no pinctrl) with a disabled mesa node in mode 1, as on J415.
    • t8122-j613.dts / t8122-j615.dts: the sep alias, plus &mesa with enable-gpios = <&pinctrl_nub 29 GPIO_ACTIVE_HIGH> and firmware-name = "apple/mesacal-<board>.bin".

    Test setup: J613, system firmware 27.0 (mBoot-20457.1.29), OS firmware 26.6.2 (mBoot-18000.161.10), the m3next Stage 1 with m1n1 aurora17-5, and the j613-25g83 GPU profile. The kernel is your 11.15 package. The test parts were apple-sep.ko rebuilt from this branch (same config and rustc 1.93.1 / bindgen 0.73.2; vermagic and every undefined symbol match the running kernel; BTF stripped) and the two J613 DTBs swapped inside the existing boot.bin. An unpatched build of those DTBs is byte-identical to the ones in your package.

    Step Result
    m1n1 with the sep alias boots. The stub provides SEPFW, lpol and ibot: iboot-manifest 6287 B, local-policy-manifest 2883 B, sep-firmware reserved
    Read-only load (xart_writes=0) SEP platform profile: T8122/J613, DART aliasing 0x0000010000004000..0x00000100ffff3fff, 12 endpoints, hwrng; keybag: missing as expected
    xart_writes=1 provision_keybag=1 sks: identity keybag provisioned (variant 5 accepted); ref-key created; xART 12 live records, 0 malformed
    Sensor power line from the device node, bound, 8000000 Hz, mode 1; the stage 0/2 registration and device-view sync pass
    Enrol enroll-completed in 6 touches; user/master catacomb state 7
    Verify right index verify-match; another finger verify-no-match
    Reboot master/owner/user catacombs Restored; Touch ID: restored 1 enrolled identity/identities; enrolment survived reboot; verify-match
    macOS afterwards Touch ID unlock still works; its two fingerprints are listed (the Linux identity isn't, as expected); adding a third worked
    Back in Linux after that catacombs Restored, identity count 1, xART 12 live records and 0 malformed (revision 17118 → 17142), verify-match

    One installer change this needs: the first enrol failed with calibration blob apple/mesacal-j613.bin could not be loaded (ENOENT). install-aurora-sep.sh had saved this Air's calibration as apple/mesa_calibration.bin, because the 11.15 J613 DTB has no sensor node. With this DT, calibration() would pick up mesacal-j613.bin from the installed DTB. For the test I copied the file across.

    Like J415, the data-ready line (AP GPIO 199, flags 3) is left out, so the driver polls (no data-ready interrupt (describe it in DT); capture falls back to SPI polling); the same follow-up as #53.

    Not tested: J615 (identical ADT values, no hardware here); lock screen and sudo through PAM; suspend and resume. Happy to open a PR against integration/aurora, or adjust first if you'd like it done differently.

    Attached (MACs, UUIDs and IPs removed): touchid.txt, which has the firmware versions, the reboot boot's SEP log and diagnostics, and the enrolment boot's SEP and fprintd log.

    touchid.txt

  2. scottjones commented on Oct 15, 2026

    @scottjones
    CollaboratorAuthor

    Drafted with an AI assistant (Claude Code); I reviewed it before posting.

    @iconidentify: J613 Touch ID after suspend: works. 13" M3 Air, 26.6.2 stub firmware, Aurora 2026.10.11.15 + the T8122 SEP profile from #82.

    • Loaded apple_sep: "restored 1 enrolled identity; enrolment survived reboot", and fprintd-verify matched.
    • systemctl suspend (s2idle) for about 2 minutes (23:27:41 → 23:29:39). On resume the sensor re-registered with identity count 1, and fprintd-verify matched again.
    • No new errors. The existing "no data-ready interrupt (describe it in DT); capture falls back to SPI polling" note is unchanged.

    Not yet tested: J615, and a 14.8.3 stub (which the current edge installer uses for M3).

  3. tonibergholm commented on Oct 16, 2026

    @tonibergholm

    Additional J613 hardware validation on 2026-10-16: Touch ID works with the 14.7 OS firmware stub, system firmware 27.0, and the published aurora-2026.10.15.19 packages. Enrollment, matching, sudo and the Omarchy lock screen passed; enrollment and matching also survived a Linux reboot.

    Tested on my 13-inch M3 MacBook Air with Codex assistance. This adds a different firmware combination and desktop authentication coverage to the results above.

    Machine and stack

    • Board/SoC: apple,j613 / apple,t8122.
    • Distribution: Arch Linux ARM, Omarchy 4.0.0.r7026.g51bfd1d-1; omarchy-mac 0.1.4-1.
    • Release: aurora-2026.10.15.19.
    • Running kernel: 7.1.12-2-20261015.19-sep-ARCH.
    • Kernel and headers packages: 7.1.12.aurora2-20261015.19.
    • m1n1-aurora 1.6.1.aurora17-5; stage2 reports v1.6.1-omarchy.aurora17. Existing stage1 reports v1.6.1-dirty and was retained.
    • asahi,os-fw-version: 14.7; asahi,system-fw-version: 27.0.
    • libfprint 1.94.100-1.1, fprintd 1.94.5-2, aurora-touchid 20261003-2.
    • Published packages used without local kernel or driver patches. Installed with --read-only --m3-handoff --no-m3-mesa first. This was a software-rendered desktop test; GPU acceleration was not enabled or qualified.

    Procedure and results

    1. Backed up the existing boot files and iBoot container. Booted the matched stack with xart_writes=0 provision_keybag=0.
    2. SEP attached as T8122/J613, advertised all 12 endpoints, and exposed /dev/sep-bio. xART diagnostics reported zero malformed or duplicate records. Sensor bound at 8 MHz, SPI mode 1. Activation stopped because no Linux identity keybag existed, as expected in this mode.
    3. Set xart_writes=1 provision_keybag=1 and rebooted. SEP provisioned the Linux identity keybag; fprintd discovered one Apple secure enclave fingerprint sensor.
    4. Enrolled one finger through fprintd-enroll, verified it, then configured the standard Aurora/Omarchy PAM paths with password fallback retained.
    5. Rebooted again and checked both live restoration and physical authentication.
    Check Result
    Calibration provisioning Pass: installer extracted this Mac's own apple/mesacal-j613.bin; no manual calibration copy needed
    Enrollment enroll-completed
    Unenrolled-finger control verify-no-match
    Enrolled finger First attempt returned verify-no-match; retry returned verify-match
    sudo sudo -k /usr/bin/true passed after fingerprint authentication; an earlier attempt timed out
    Omarchy lock screen Fingerprint PAM logged successful authentication; session returned to an unlocked, responsive desktop
    Graphical authorization pkexec /usr/bin/true completed after the graphical authentication prompt
    Enrollment after Linux reboot Master, owner and user Catacombs restored; one identity present in SEP and fprintd
    Physical match after reboot fprintd-verify returned verify-match
    sudo after reboot sudo -k /usr/bin/true passed after fingerprint authentication

    Relevant post-reboot kernel messages:

    sbio: cold-match preparation true, component restore true
    Touch ID: restored 1 enrolled identity/identities; enrolment survived reboot
    sbio: restore-ready after device-view proof true
    

    Remaining limits and documentation

    Not tested on this machine: macOS Touch ID after Linux enrollment, Linux/macOS round-trip persistence, suspend/resume, or physical password-fallback authentication. Password paths remain configured. These results do not qualify J615 or other M3 models.

    The matched installer's final summary still says "Touch ID is not supported on M3 yet." That appears stale for this J613 configuration. Would you prefer a small documentation/installer-message PR, scoped to the tested board and firmware combinations?

    No calibration, keybags, raw device trees, biometric payloads or disk images are attached.

  4. scottjones commented on Oct 23, 2026

    @scottjones
    CollaboratorAuthor

    Drafted with an AI assistant (Claude Code); I reviewed it before posting.

    J613 (13" M3 Air), reinstalled: Touch ID bring-up wedges the enclave after the xART guard refuses a record the earlier install left.

    Setup

    • Hardware and boot: MacBook Air 13" M3 (J613). A fresh install from 2026-10-21 (installer 2.3.5, 14.x stub, os-fw 14.7). The boot chain is aura 0.1.0.r20261023, and m1n1 forwards /chosen/apfs-preboot-uuid.
    • Kernel: 7.2.11 (linux-aurora 7.2.11.aurora1, run here as a local 7.2.11-m3air build). SEP platform profile: T8122/J613.
    • The driver was loaded by omarchy-mac's consent flow (Offer Touch ID as a fingerprint reader once its owner agrees omarchy-mac-pkgs#12): modprobe apple-sep with xart_writes=1 provision_keybag=1, after extract-mesa-calibration wrote apple/mesacal-j613.bin.
    • No Linux key store existed on this install before.
    • Earlier on this Mac: its previous install, on the 26.6.2 stub, enrolled and matched through this driver on 2026-10-11, and after suspend on 2026-10-14 (comments above). This is the first try since the reinstall.

    What happened

    apple_sep 29e400000.sep: SEP platform profile: T8122/J613
    apple_sep: xART guard level 1: 9 session records exist, 0 created by Linux; the other 9 are protected
    apple_sep 29e400000.sep: xART: APFS-resolved .gl extent (base 0x4d4000); 170 slots, 11 live records, max revision 17617, 0 malformed, 0 duplicate, 0 repaired; writes ENABLED
    apple_sep: created a new host-state store
    ...  (endpoints 0x14 0x18 0x00 0x0e 0x10 0x13 0x15 0x17 0x08 0x0c 0x0a 0x12 discovered normally)
    apple_sep 29e400000.sep: xART: using /chosen/apfs-preboot-uuid
    apple_sep: xART guard: refused to rewrite a kind-4 record this Linux install did not create
    apple_sep 29e400000.sep: sks: CREATE_KEYBAG got no reply after 2061 ms
    apple_sep 29e400000.sep: sks: CREATE_KEYBAG exchange did not complete (buffers unregistered, wedged, or OOL write failed)
    apple_sep 29e400000.sep: bringup: identity-keybag provisioning failed; no retry this boot
    apple_sep 29e400000.sep: hwrng: survey attempt 1 failed (ETIMEDOUT); retrying
    ...  (attempts 2-4)
    apple_sep 29e400000.sep: hwrng: survey failed after 5 attempts (ETIMEDOUT)
    
    • diag read touchid=failed sensor_attach=unknown restore=unknown matching=blocked xart=enabled.
    • /dev/sep-bio never appeared.
    • The enclave stopped answering for the rest of the boot. A reboot without loading the driver brought it back.

    What's different from the working runs

    @tonibergholm's J613 above works on the same 14.7 stub. That Mac differs from this one in two ways:

    • Kernel: it ran the published 7.1.12 -sep kernel (aurora-2026.10.15.19). This Air runs 7.2.11.
    • History: that Mac had no earlier Linux Touch ID setup. This Air did: its previous install, 2026-10-11 to 10-20, enrolled through this driver and was then removed for a reinstall.

    So the 14.7 stub doesn't look like the cause.

    Questions

    1. The record and the hang. Is the refused kind-4 record the previous Linux install's? This install has a new preboot UUID, so it isn't the install that created it. The guard was right to refuse, but CREATE_KEYBAG then hung the enclave instead of failing back. Could the driver stop before CREATE_KEYBAG when the guard refuses, and report why?
    2. Recovery. What should a reinstalled Mac do? Is there a safe way for the new install to start its own lane, or does the old record have to be cleared, for example from macOS?
    Full apple_sep kernel log from the failed boot (no serials)
    [298.074030] apple_sep: apple_sep: SEP DART state: enabled=false, apple,dma-offset present=true, apple,dma-range present=true
    [298.075012] apple_sep: apple_sep: SEP DART changeset applied (offset BIT(40), range 0..4 GiB, status okay)
    [298.075346] apple_sep: apple_sep: SEP node enabled; platform device should now exist
    [298.075366] apple_sep 29e400000.sep: SEP platform profile: T8122/J613
    [298.082004] apple_sep: xART guard level 1: 9 session records exist, 0 created by Linux; the other 9 are protected
    [298.082028] apple_sep 29e400000.sep: xART: APFS-resolved .gl extent (base 0x4d4000); 170 slots, 11 live records, max revision 17617, 0 malformed, 0 duplicate, 0 repaired; writes ENABLED
    [298.083006] apple_sep: created a new host-state store
    [298.083020] apple_sep 29e400000.sep: trusted-keys: registering "trusted" key type backed by the SEP ref-key seal
    [298.362059] apple_sep 29e400000.sep: discover: endpoint 0x14 (type 0, msg0 0x68696265140000fd); 1 known
    [298.362353] apple_sep 29e400000.sep: discover: endpoint 0x14 (type 1, msg0 0x01010101140100fd); 1 known
    [298.362607] apple_sep 29e400000.sep: discover: endpoint 0x18 (type 0, msg0 0x73746163180000fd); 2 known
    [298.362863] apple_sep 29e400000.sep: discover: endpoint 0x18 (type 1, msg0 0x02020202180100fd); 2 known
    [298.363135] apple_sep 29e400000.sep: discover: endpoint 0x00 (type 0, msg0 0x636e746c000000fd); 3 known
    [298.363378] apple_sep 29e400000.sep: discover: endpoint 0x00 (type 1, msg0 0x00000000000100fd); 3 known
    [298.364019] apple_sep 29e400000.sep: discover: endpoint 0x0e (type 0, msg0 0x686463700e0000fd); 4 known
    [298.364300] apple_sep 29e400000.sep: discover: endpoint 0x0e (type 1, msg0 0x010101010e0100fd); 4 known
    [298.367022] apple_sep 29e400000.sep: discover: endpoint 0x10 (type 0, msg0 0x78617273100000fd); 5 known
    [298.367294] apple_sep 29e400000.sep: discover: endpoint 0x10 (type 1, msg0 0x02020202100100fd); 5 known
    [298.367554] apple_sep 29e400000.sep: discover: endpoint 0x13 (type 0, msg0 0x7861726d130000fd); 6 known
    [298.367797] apple_sep 29e400000.sep: discover: endpoint 0x13 (type 1, msg0 0x02020202130100fd); 6 known
    [298.368062] apple_sep 29e400000.sep: discover: endpoint 0x15 (type 0, msg0 0x706e6f6e150000fd); 7 known
    [298.368320] apple_sep 29e400000.sep: discover: endpoint 0x15 (type 1, msg0 0x05050707150100fd); 7 known
    [298.913017] apple_sep 29e400000.sep: attach: 7 endpoints advertised in 18 messages
    [298.913111] apple_sep 29e400000.sep: xART: using /chosen/apfs-preboot-uuid
    [300.465020] apple_sep 29e400000.sep: discover: endpoint 0x17 (type 0, msg0 0x736b646c170000fd); 8 known
    [300.465311] apple_sep 29e400000.sep: discover: endpoint 0x17 (type 1, msg0 0x01010101170100fd); 8 known
    [300.465560] apple_sep 29e400000.sep: discover: endpoint 0x08 (type 0, msg0 0x7362696f080000fd); 9 known
    [300.465810] apple_sep 29e400000.sep: discover: endpoint 0x08 (type 1, msg0 0x01010101080100fd); 9 known
    [300.466084] apple_sep 29e400000.sep: discover: endpoint 0x0c (type 0, msg0 0x737365200c0000fd); 10 known
    [300.466333] apple_sep 29e400000.sep: discover: endpoint 0x0c (type 1, msg0 0x010101010c0100fd); 10 known
    [300.476025] apple_sep 29e400000.sep: discover: endpoint 0x0a (type 0, msg0 0x736372640a0000fd); 11 known
    [300.476291] apple_sep 29e400000.sep: discover: endpoint 0x0a (type 1, msg0 0x010101010a0100fd); 11 known
    [300.685022] apple_sep 29e400000.sep: discover: endpoint 0x12 (type 0, msg0 0x736b7320120000fd); 12 known
    [300.685292] apple_sep 29e400000.sep: discover: endpoint 0x12 (type 1, msg0 0x02010201120100fd); 12 known
    [301.486023] apple_sep: xART guard: refused to rewrite a kind-4 record this Linux install did not create
    [303.265025] apple_sep 29e400000.sep: sks: CREATE_KEYBAG got no reply after 2061 ms
    [303.265259] apple_sep 29e400000.sep: sks: CREATE_KEYBAG exchange did not complete (buffers unregistered, wedged, or OOL write failed)
    [303.265441] apple_sep 29e400000.sep: bringup: identity-keybag provisioning failed; no retry this boot
    [305.313024] apple_sep 29e400000.sep: hwrng: survey attempt 1 failed (ETIMEDOUT); retrying
    [308.385030] apple_sep 29e400000.sep: hwrng: survey attempt 2 failed (ETIMEDOUT); retrying
    [312.481020] apple_sep 29e400000.sep: hwrng: survey attempt 3 failed (ETIMEDOUT); retrying
    [318.689028] apple_sep 29e400000.sep: hwrng: survey attempt 4 failed (ETIMEDOUT); retrying
    [328.865028] apple_sep 29e400000.sep: hwrng: survey failed after 5 attempts (ETIMEDOUT)
    

    🤖 Generated with Claude Code

  5. satchlj commented on Oct 26, 2026

    @satchlj

    Additional J613 Touch ID validation on 2026-10-25: one saved fingerprint survived a normal Linux reboot and unlocked the ordinary Omarchy lock screen. The owner physically confirmed password login/unlock, typing and trackpad movement after the final reboot as well.

    This is a local research build, not qualification of the latest published Aurora kernel and not a proposed fix for the reinstall/foreign-session-record failure in the latest comment above.

    Machine and stack:

    • 13-inch M3 MacBook Air, apple,j613 / apple,t8122, Arch Linux ARM/Omarchy.
    • Current DT reports OS firmware 26.6.2, system firmware unknown. I am reporting the current readback rather than substituting an earlier firmware value.
    • Running kernel 7.1.9-j613-4, local package linux-j613 1:7.1.9.j613.4-11; m1n1-aurora 1.6.1.aurora1-3; Omarchy 4.0.0.r7025.g8a787b6-1; omarchy-mac 0.1.4-1.
    • Local SEP module SHA-256 f6cfe0a4e88bec9abf54d9e8b4bfe1725916dae7df887a3d59198b353e8f0db1; this includes local recovery/diagnostic work and is not asserted identical to the current public driver.
    • libfprint 1.94.100-1.2, fprintd 1.94.5-2.1, normal Quickshell fingerprint PAM. The vendor fprintd sandbox remains enabled.

    Results:

    Check Result
    One attended enrollment and durable saved state Pass
    Reboot, saved finger match, different unenrolled finger rejection Pass
    Ordinary lock-screen password unlock Pass, owner physically tested
    Ordinary lock-screen fingerprint unlock Pass, owner physically tested
    Final normal reboot, automatic loader/fprintd startup, then ordinary fingerprint unlock Pass
    Keyboard, trackpad and password login after that reboot Pass, owner physically tested
    J615, current public release, suspend/resume, new macOS round trip Not tested in this window

    The daily existing-bag policy explicitly uses xart_writes=1 provision_keybag=0 xart_guard=1 trusted_keys=0 filevault_api=0 fresh_enrolment_only=0. The earlier local guard-2 experiment prevented the shared-root persistence updates; normal guard 1 works here. Current public M3 defaults already select guard 1, so this observation does not justify changing upstream to an unguarded mode.

    The final boot was observed without any manual module/daemon start: the loader ran at 4.56–7.87 seconds and fprintd started at 13.70–14.23 seconds, then idle-exited/reactivated normally. All one-shot recovery/test packages have now been removed; a dependency-only normal-service package update leaves its five working runtime files byte-identical.

    Unused bags from the earlier attempts remain, and safe deletion is not established. Backups and spent-authority receipts are retained privately. No raw xART replay, anti-replay counter reset, key deletion or forced internal save was used. No enrolled data, identifiers, calibration bytes, device trees, keybags or whole journals are attached.

    Prepared with Codex assistance from the owner-confirmed results and read-only final boot/package observations.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions