Skip to content

arm64/sep: describe J813 Touch ID and admit its T8142 profile - #1732

Draft
DjDeveloperr wants to merge 9 commits into
integration/aurorafrom
review/j813-touchid-70035185bd9c
Draft

DjDeveloperr wants to merge 9 commits into
integration/aurorafrom
review/j813-touchid-70035185bd9c

Conversation

@DjDeveloperr

@DjDeveloperr DjDeveloperr commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

Current draft head 29991004b29bd84a00b7a4fc919fb0dce1f14016 (tree c753e785a63e7a979a663b7111932cc8453b7f94). Merged integration at fc9c78fa after #2135. Integration now supplies the J813 SEP nodes, high DMA window and board admission. This branch retains the fingerprint SPI2 node, sensor GPIO reservations and SEP profile test coverage as a two-file change. The baseline and updated J813 DTBs compile locally with the same DTC warnings. The updated DTB keeps the SEP alias and contains the SPI2 fingerprint child. All eight CI checks passed on synthetic merge aa94bab5, whose tree matches this head. Host tests passed 425/425. Device tree checks built 137 DTBs on each side with zero new warnings. SEP checks passed, including 306/306 KUnit tests. Aura #165 remains draft and unmerged. Existing loader qualification, adoption and SEP policy requirements still apply.

J813 hardware follow-up. Owner report tested this PR’s J813 DTS change on the #2135 + #2142 kernel with a 27.2 / 26B123 stub and EL2 Aura 8b9c92ea. The composition omitted this PR’s profile naming and tests. With CONFIG_SPI_APPLE=m, SPI2 bound and the fingerprint sensor appeared. The owner reports successful per-device calibration, interrupt-driven capture on AP GPIO 200, fprintd enrollment and matching, and enrollment restoration across reboots. The reported guard setting was xart_guard=1, with zero refusals and zero xarm failures in the reported run.

Resume limit. Deep sleep resumed and the sensor returned online, but captures failed until reboot. The owner reports successful post-wake capture and swaylock unlock with the separate SEP recalibration fix #2202. That fix is still open. This result covers the DTS change on the reported composition. Existing loader qualification and adoption remain pending.

Previous heads and validation history

Current draft head 3a3d87b162f24e58c0943bdf321ebfb1be86d3ae (tree 4b8ee8c4fc65c04725817f6789c65755da042e78). Updated from integration after the backlight fixture fix #2168 landed as 432f1758. The PR still changes only the J813 device tree and SEP profile. All eight checks passed on synthetic merge b2ec4e60, whose tree matches this head. Host tests passed 425/425, including the backlight fixture. DT checks built 137 DTBs on each side with zero new warnings. SEP checks passed, including 306/306 KUnit cases. Aura #165 remains draft and unmerged. Existing loader qualification, adoption and SEP policy requirements remain.

Previous head and validation history

Current draft head 4a05fb9975b0fcef4793035b733923f31870b899 (tree 5447efb23343f1417c9c4d30619ecb67a49a45a6). Parent #1532 has landed. This branch now includes integration/aurora@7f0a1d27 through a normal merge. The remaining diff against that base is the J813 DTS and SEP profile. The shared sensor test keeps main’s J716c case alongside J813 and J815, for 18 cases. Compared the two-file feature delta against the previous version and checked whitespace. Host CI passed 423/425 cases. Both failures are compile errors in the inherited backlight fixture. They reproduce on integration without the Touch ID changes. The separate fixture fix #2168 passed the local backlight suite and all five negative controls. Aura #165 loader qualification and adoption, along with the existing hardware and SEP policy limits below, still apply.

Previous head and validation history

Draft at b5941ec68bdad4e54729d8dfb7a3be79ffd58aab (tree daca0767db4dfebe01ec99e2533982b251232744), stacked on parent#1532@c93ee625. Do not merge before the parent.

Normal merge of preceding1008ae4f with the updated parent/current main. The J813 board and SEP profile feature files are byte-identical to their preceding versions; their net follow-up remains two files. Both board-layer comparisons pass and the SEP profile's net edits remain exact. The parent GPU-schema merge retains T8142's seven-region ABI and main's G17 rules.

Current-head Host43560809639/job132343834251 passed323/323 runs. All six SEP43560809632 jobs passed: source, Rust module build, three DWC3/ATC configurations and KUnit. KUnit132343834349 executed306 tests, all passed, including t8142_sensor_boards_require_the_matching_soc. Seven non-DT checks are successful on this exact head.

DT43560809641/job132343834271 built137 baseline and137 head DTBs (make return0), with399→391 warning lines: one new and nine removed. The sole new warning is the inherited J815 disp0_piodma alias. Seven changed bindings—including the merged GPU binding—and ADT's19 tests/six coverage comparisons passed without regression. No new SEP/pin/GPU/PMU/MCC/AVE warning appeared. The overall DT check remains failed, not green. No unchanged-source retry, warning filter or alias rename was made. Earlier323-host/306-KUnit/all-six-SEP successes remain historical for1008ae4f. Effective admission/write/guard semantics and the owner's explicitly configured guard1 evidence keep their existing scope; no new runtime-default, rejection, IRQ-only, multifinger or sleep qualification is claimed.

Keep parent-first landing, affected Aura#165 loader qualification/adoption, alias migration, final checks/review and existing physical/policy limits. No XART/enrolment/user-state, hardware, media or loader/default operation occurred.

Retained previous description and validation history (older heads)

Draft at 1008ae4ff43ff310e3a0d3400975415a4e4a1d4c (tree c7bd8c936064a70d60f62c21ad3493c3322d8be3), stacked on parent#1532@a244cd1d. Do not merge before the parent.

Normal merge of old98e509f0 with the updated parent, retaining original ancestry. Carries current-main PMU/AVE prerequisites and @gig3m's supplied J813 PMU/MCC/AVE patch. The J813 EOF overlap retains both the complete SEP block and the new board enables. The net follow-up remains two files: J813 DTS and the SEP profile. Four symbolic board/layer comparisons pass; profile added/removed lines match the original feature delta exactly.

This update makes no new SEP policy change. Existing admission/default-write/guard semantics and the owner's explicitly configured guard1 evidence retain their documented scope below; no runtime inheritance, rejection, multifinger, IRQ-only or sleep qualification is added. The PMU/cache/encoder report is not SEP qualification.

Current-head Host43541172933/job132281828593 passed323/323 runs. All six jobs of SEP43541172944 passed: source, Rust module build, three DWC3/ATC configurations and KUnit. KUnit job132281829211 executed306 tests, all passed, including t8142_sensor_boards_require_the_matching_soc. Together, seven non-DT checks are successful on this exact head.

DT43541173041/job132281829282 built137 baseline and137 head DTBs (make return0), with399→391 warning lines: one new and nine removed. The sole new warning is the inherited J815 disp0_piodma alias. Seven changed bindings and ADT's19 tests/six coverage comparisons passed without regression; no new J813 SEP/pin/PMU/MCC/AVE warning appeared. The overall DT check remains failed. No unchanged-source retry, warning filter or alias rename was made. Old native/QEMU results remain historical for98e509f0, not this new source. Parent-first landing, affected Aura#165 loader qualification/adoption, alias migration, final checks/review and physical/policy limits remain. No XART/enrolment/user-state, hardware, media or loader/default operation occurred.

Retained prior description and validation history (older source identities)

Draft at 98e509f06b380cb69c91d38d1e74dde473dc208d, stacked on #1532 at0434a61e. Do not merge before that parent. The net follow-up remains two files. This normal merge carries the updated parent/current main while preserving the original SEP/profile feature patch exactly.

Carry @gig3m’s 6837017905 patch: J813 SEP/mailbox/SPI2/Mesa/DART nodes with board-specific calibration filenames, loader SEP alias, AP200/NUB36 exposure while retaining audio pins, and J813 admission only with T8142. SEP/DART remain initially disabled for the driver’s existing changesets.

The diagnostic label names both Airs. Eight selector cases include J704 refusal; both Airs are in known-sensor coverage, T8142 in the BIT(40) window check, with a guard1 assertion. Ryan’s5883010555 review accepts the bounded admission at the J815 precedent, subject to parent-first landing and final CI; optional nits were addressed. It is COMMENTED, not formal approval. The separate aurora-xart model-list gaps are accepted as a utility follow-up.

Effective policy: admitting J813 makes it eligible for the existing xart_writes=1/provision_keybag=1 defaults. Parameter xart_guard=-1 inherits level1. This feature does not change those defaults or guards and is not read-only by default. The owner explicitly configured guard1; runtime inheritance from-1 was not tested.

Owner evidence is the223d feature on92195659, J813/Mac17,3/27.0 stub, Aura1.0.0b12/aurora-sep1-25: read-only attach, then owner-enabled keybag provisioning, single-finger enrol/lock-screen match, working macOS Touch ID, and restored/matching Linux identity after reboot. Follow-up6837844779 reports a later third write-enabled boot with the list loaded,0 refusals/0 XARM failures, keybag present and identity restored; current matching is unverified. Ten session records are protected and0 recorded Linux-created. This does not test rejection paths, IRQ-only#1596 operation, multifinger, lockout or deep sleep, and is not a boot of this new whole tree.

The parent integration preserves its J815 graph/MCC and J813 N1/ALS work, including main RTC changes; MCC is inherited disabled on J813. The duplicate G17s schema enum repair is inherited from the parent. Six symbolic board/layer checks, eight metaschema/reference checks, five T8142 memory cases and560 non-target comparisons pass. Three actual KUnit bodies pass in a host Rust extraction on the new source.

Current98e509f0 Host43379009670/job131732157206, arm64 SEP module131732156680 and source checks131732157091 passed their substantive steps. Arm64 QEMU KUnit131732156983 ran295 tests, all passed; the log explicitly includes the T8142 board/guard, shared known-sensor and BIT(40) window tests. All three DWC3/ATC compile profiles now passed as well (131732157105/7117/7133); all six jobs of SEP43379009612 completed successfully. Together with Host, seven software checks pass; DT remains the inherited-alias failure below.

DT43379009604/job131732156929 built137 DTBs on both base and head (make rc=0/0),402/394 warning lines, one new/nine removed. All seven changed bindings—including the repaired AGX schema—and ADT19/six baseline comparisons passed with no regression. The sole new warning is the inherited J815 disp0_piodma alias; no new J813 SEP/pin/MCC/RTC/ALS warning appeared. The overall DT check remains failed. No unchanged-source retry or warning filter was used. Aura165 remains draft without a new posted qualification/adoption result; no hardware/default-policy qualification is inferred from these software results.

Remaining: parent1532 must land after Aura165’s affected-loader qualification/adoption and coordinated alias migration, then final required checks/reviews must pass. Keep draft. No warning filter, unchanged-source retry, physical test, pairing/enrolment/XART write, user-state or loader/default action was performed by this review.

Based on @gig3m’s supplied patch/report (assisted by Claude Fable5.2); prepared and integrated with LLM assistance from Codex. Original ancestry retained.

Carry the J813 owner's SEP/mailbox/SPI/Mesa/DART block and board-specific
calibration filenames. Add the SEP alias and expose AP GPIO200/NUB GPIO36
while preserving the audio pins. Admit J813 only alongside the existing
J815 T8142 selector; retain other board and SoC admission boundaries.

Based on @gig3m's supplied patch and owner report in linux#1532 comment
6837017905. Add KUnit selector cases for both accepted boards and unknown,
missing or mismatched SoC/board identities. Local command creation failed
with EAGAIN; these new cases require the hosted Rust/QEMU checks.

This changes effective board admission: existing xart_writes=1 and
provision_keybag=1 defaults remain, with profile-inherited guard level1.
It is not read-only by default. Keep this follow-up draft pending SEP/XART
policy review, CI and a decision on the separate aurora-xart model-reporting
gap. No defaults, guard definitions, APFS checks or user state are changed.

The reported single-finger/read-only-then-write/reboot/macOS checks do not
qualify IRQ-only capture, multifinger, lockout or deep sleep. No hardware,
XART write, enrolment, provisioning or loader action was executed here.

Assisted-by: LLM

@RyanTheTide RyanTheTide left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewed the e8fb89d...7003518 delta.

The J813 sep, mailbox, spi2, Mesa and sep_dart nodes match J815's except for the calibration file names. The pinctrl overrides come after t8142-j813-audio.dtsi, keep every pin it exposes, and add only AP 200 and NUB 36. The selector admits apple,j813 only with apple,t8142; the other branches and the T8142 profile are unchanged, so no other board's profile, guard or defaults move.

The J813 run (read-only attach, keybag, enrol, lock-screen match, macOS Touch ID afterwards, enrolment restored after reboot, guard 1, no refusals) covers what #1531 admitted J815 on.

Before merge:

  • #1532 lands first and this delta is still these two files.
  • kunit, rust-build, dt-checks and host-tests pass on the final head.

Fine here or later:

  • Rename the label to "T8142/J813, J815", as T6041 was for J614c. Only the probe log and diag print it.
  • Add apple,j704 (a T8142 board in the tree) as a negative case, J813/J815 to every_touch_id_mac_gets_a_sensor_the_shared_path_decodes and the BIT(40) window test, and assert T8142.xart_guard == 1.
  • Could you post the xart_guard, xart_guard_list and refusal diag values from after the reboot? The report says the list stayed empty after a keybag create and an enrolment.
  • aurora-xart's platform.py lacks T8142, J714s, J614c and J715. Separate aurora-xart PR; the driver's diag still reports the real guard level.

@gig3m

gig3m commented Nov 12, 2026

Copy link
Copy Markdown
Contributor

The values asked for, from J813 on the current boot (the third write-enabled boot since the enrolment; macOS was booted in between):

xart: enabled
xart_guard: 1
xart_guard_list: loaded
xart_guard_refusals: 0
xarm_failures: 0
keybag: present
restore: restored
matching: unverified
policy: xart_writes=1 provision_keybag=1 xart_guard=1 trusted_keys=1 filevault_api=1 xarm_trace=0 refkey_v2=0 identity_override=0 identity_source=dt
apple_sep: xART guard level 1: 10 session records exist, 0 created by Linux; the other 10 are protected
apple_sep 30a600000.sep: xART: APFS-resolved .gl extent (base 0x4df000); 170 slots, 12 live records, max revision 1965, 0 malformed, 0 duplicate, 0 repaired; writes ENABLED
apple_sep 30a600000.sep: Touch ID: restored 1 enrolled identity/identities; enrolment survived reboot

/var/lib/aurora-sep-xart-owned.bin is 0 bytes and aurora-xart status shows Guard list 0 records (0 bytes), Keybag slot committed, read back from the bag UUID (1612 bytes, checksum ok).

The guard line read the same (10 session records, 0 created by Linux) on the boot that provisioned the keybag and on every boot since. Live records stayed at 12 while max revision went 1924 (read-only boot and the provisioning boot) -> 1961 (first boot after macOS) -> 1965 (now).

One correction to my report: xart_guard=1 was set explicitly in /etc/modprobe.d/aurora-sep.conf here, not left at -1.

Assisted-by: Claude Fable 5.2

Inherit the owner-provided J813 ALS node without changing the existing
audio graph or SEP profile/policy delta. Preserve both histories.
Fresh CI and the independent parent/loader/review gates remain.

Assisted-by: LLM (Codex)
Address Ryan's review5883010555 of the J813 follow-up. Name both J813
and J815 in the diagnostic profile label. Add J704 as a negative board,
both Airs to the shared known-sensor test, T8142 to the BIT40 window
test and an explicit level-one guard assertion. Compare the selected
profile's stable name rather than promoted-constant pointer identity.

The production selector is unchanged. Apart from the diagnostic label,
T8142 profile fields, geometry, guard and defaults are exact previous
source. The reviewer accepts aurora-xart model coverage as a separate
utility PR; do not change it or user state here. The owner corrected
the report: guard1 was explicitly configured, not inherited at runtime.

Three actual KUnit bodies pass in a host Rust extraction using actual
profile definitions, selector and exact-head known-sensor IDs. Host
CStr/error/static-assert glue is used; this is not a kernel module,
arm64 QEMU KUnit or physical default-policy test. Fresh hosted checks
and the parent1532/loader-alias gate remain before merge.

Assisted-by: LLM (Codex)

@DjDeveloperr DjDeveloperr left a comment

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Addressed Ryan’s review5883010555 in8cc28060. The diagnostic label now names J813 and J815. Added J704 as a refusal case, both Airs to the known-sensor coverage, T8142 to the BIT(40) window check and an explicit guard1 assertion. The selector test compares the stable selected profile name instead of promoted-constant pointer identity. Apart from that diagnostic label, every T8142 runtime field and the production selector are byte-exact previous source; the follow-up remains two files relative to parent09e6fbbc.

Three actual KUnit bodies pass in a host Rust extraction using the actual profile constants, selector and exact-head known-sensor IDs. The host uses CStr/error/static-assert glue; it is not a kernel Rust/module build, arm64 QEMU KUnit or runtime-default qualification. Initial exact-blob and extraction-helper failures were corrected without changing production behavior; logs are retained. Fresh current-head hosted CI is pending/queued.

The requested owner diagnostics are now posted in6837844779: explicit guard1, loaded guard list,0 refusals/0 XARM failures, keybag present and identity restored after the third write-enabled boot with macOS between. Ten session records remain protected,0 are recorded Linux-created, and the owned list is0 bytes. The current diagnostic says matching unverified; the previous lock-screen matching result is an earlier observation. This does not test rejection paths or the inherited -1 default. No new hardware, enrolment, XART/user-state or loader action was performed here.

Following Ryan’s stated scope, aurora-xart model coverage stays a separate utility PR. Keep this draft until1532 lands and final-head kunit/rust-build/DT/Host checks pass; the parent’s Aura165 affected-loader qualification/adoption and alias migration are still outstanding. Ryan’s review is COMMENTED, not relabelled as a formal approval.

Retain the full audio or SEP feature while inheriting the updated
parent/main, complete MCC block and semantics-preserving duplicate
enum repair. Keep both published histories and all review/loader
qualification gates; require fresh CI.

Assisted-by: LLM (Codex)
Carry the current-main prerequisites and @gig3m's tested J813 PMU/MCC/AVE
enablement. Preserve the existing SEP/profile feature delta and
all current shared definitions. Resolve the J813 EOF overlap by retaining both the SEP block and the new
PMU/cache/encoder enables.

Symbolic three-way property checks preserve both board descriptions and
the parent enablements. The net SEP profile edits remain exact.
Existing loader-alias and physical/policy qualification gates remain.
No hardware, audio playback, XART write or loader/default change occurred.

Assisted-by: LLM
Signed-off-by: Diljit Singh <hey@djdev.me>
Retain the original feature commits and pull in the current-main GPU
binding merge. The J813 board and SEP profile feature files remain byte-identical
and four symbolic board/layer comparisons preserve both inputs.

Existing parent-first, loader-alias and physical/policy gates remain.
No new kernel/DTB/hardware or loader/default operation was performed.

Assisted-by: LLM
Signed-off-by: Diljit Singh <hey@djdev.me>
Keep the J716c shared SEP test entry from main alongside the existing J813 and J815 cases. Retain the two-file Touch ID delta without changing SEP policy.

Signed-off-by: Diljit Singh <hey@djdev.me>
Use integration’s existing J813 SEP nodes, high DMA window and board
admission. Retain the fingerprint SPI node, GPIO reservations and
profile tests without changing the SEP guard policy.

Signed-off-by: Diljit Singh <hey@djdev.me>
@sithglan

Copy link
Copy Markdown
Contributor

Tested the J813 part on my J813 (MacBook Air 13" M5, macOS 27.2 / 26B123 stub, EL2 aura 8b9c92ea). I took this PR's 50-line t8142-j813.dts change and put it on top of my consolidated J813 kernel (#2135 plus #2142). The profile.rs change is naming and tests only, so I left it out.

  • SPI: my config had CONFIG_SPI_APPLE off. With it on as a module, apple-spi binds 3a5108000.spi and the sensor appears as spi1.0 (spi:mesa-fingerprint).
  • Calibration: aurora-touchid-setup built apple/mesacal-j813.bin and apple/msrk-j813.bin from this Mac's factory records, with the seal naming this Mac. The names come from the node's firmware-name.
  • SEP: driver defaults with xart_guard=1. The first write-enabled boot provisioned the identity keybag; the guard allowed its one rewrite of a pre-existing kind-4 record and refused nothing afterwards (0 refusals, 0 xarm failures). 10 session records exist, 0 created by Linux.
  • Touch ID: the sensor came online with interrupt-driven capture on AP GPIO 200 (apple-mesa-drdy). Enrolment with fprintd (libfprint 1.94.100 aurora13) worked, and so did matching. The enrolment survived reboots ("restored 1 enrolled identity").
  • Deep sleep: with the sensor bound, deep sleeps work and the sensor comes back online. But no capture saw a finger after a deep sleep, until the next boot, because the enclave loses the sensor calibration while it sleeps. soc: apple: sep: calibrate the Touch ID sensor again after a system sleep #2202 fixes that. With it, the first capture after each wake scans, and swaylock unlocked with a touch after a deep sleep.

The DTB with this node has been my default boot since today.

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

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants