Skip to content

omamobile-fingerprint 0.1.17, device-panther: fingerprint on the Pixel 7 - #458

Closed
angelrenefm wants to merge 10 commits into
derekross:mainfrom
angelrenefm:panther-fingerprint-pr
Closed

angelrenefm wants to merge 10 commits into
derekross:mainfrom
angelrenefm:panther-fingerprint-pr

Conversation

@angelrenefm

Copy link
Copy Markdown
Contributor

Fingerprint unlock works on my Pixel 7: a right index finger enrolled from the shell's enrollment screen, and the lock screen unlocked with it, also after a reboot with omamobile-fp enabled and this PR's blacklist in place (Trusty loaded from the DT, the unit started the stack by itself, no failed units).

  • omamobile-fingerprint knows the Pixel 7 (props-panther.conf, its touchscreen), allows local HBM on its S6E3FC3 panel, and builds the bridge with the NDK (pipe2 needs _GNU_SOURCE).
  • The TDP guard takes an empty persist:/ss/0. On this phone it is a 0-byte file with the factory date (1 Jan 2020), the same in the 11-03 and 11-04 backups: its Trusty never committed TDP, and stock's storageproxyd serves that empty file on every boot (init.gs201.rc links /data/vendor/ss/persist to persist's /ss). The runner refused it as it refuses a missing one. Now an empty regular file is taken as an empty TDP copy and its hash recorded; a missing or unreadable /ss/0 is still refused, and so is any later change to the copy or to persist:/ss/0. test/preflight covers it and fails with the old runner. After the first start (11-04), two days and one enrollment later (11-06), the copy was still the empty file: Trusty did not write TDP.
  • device-panther blacklists sg and goodix_fp, as the 7 Pro does, so omamobile-fp is what loads them. Not Trusty, unlike the 7 Pro's file: there the sound start (omamobile-aoc-start) loads it, and this package has none, so the trusty_* lines would leave omamobile-fp without /dev/trusty-ipc-dev0. Trusty loads from the DT's /trusty node, as on my phone.
  • device/panther/FINGERPRINT-PLAN.md: gates P1-P5 with results.

Tested on: kernel 7.3.0-rc6-omamobile-push1-g44048a9aa795 with the fingerprint DT commit (LDO15S disabled, switched on by the fingerprint-sensor overlay) and the LHBM panel commit, both still to be proposed in linux-omarchy-mobile; Google's files from the phone's own image (extract-fingerprint-vendor -d panther, CP3A.261005.002.A1). The HAL put the sensor at (540, 1713), radius 100, and reported no calibration error.

Not checked:

  • the packages installed with pacman -U: their files ran copied by hand (the blacklist is this PR's file);
  • a wrong finger, and sudo through pam_fprintd;
  • stock Android after a factory reset (cheetah's T2).

Version: omamobile-fingerprint is 0.1.17 because #440 claims 0.1.16; device-panther's number is yours to set.

I used AI help (Claude) to work on this.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @angelrenefm, this is a big one done carefully: P1 to P5 in order, read-only first, your own "yes" before the irreversible step, and an enrolment that survives a reboot with the unit starting the stack by itself. Taking the 7 Pro's TA, HAL and runner as they are, with per-device values only where the Pixel 7 really differs (panel, touchscreen), is the right shape. CI passes, and I ran the runner's preflight test, the fingerprint package test and the device package test on your branch merged onto today's main: all pass.

The part I need to decide myself is the empty persist:/ss/0. Your reasoning convinces me: stock has served that empty file since the factory, so serving the same empty file is what stock does, and any change on either side still stops the start. But it is a change to the guard that protects Trusty's storage, and our rule is that those guards don't change without my say-so. I'm inclined to say yes, with one addition:

  1. A test for the size check. I deleted the Size: 0 line in persist_dump_tdp() and the preflight test still passed, and the same for the Type: regular line. The size check is what stops a non-empty /ss/0 whose dump fails (a read error) from being taken as empty at the first start. That would hand Trusty an empty TDP while persist holds a real one. Could you add a case where the fake debugfs's dump produces nothing while stat reports a regular file of, say, 4096 bytes (it must refuse, with no copy and no record), and one where stat reports something that isn't a regular file?
  2. The plan's status line still says nothing has run on the phone, while section 6 has P1-P5 passed. Could you update it, point at linux-omarchy-mobile#207 where it names your local branch and commit, and fix the section numbering (there's no section 5)?

Smaller, for later rather than now:

  • Trusty loads straight from the DT on the Pixel 7, while on the 7 Pro it's loaded by a start script behind a hang marker, because it calls into the secure world at probe. That's fine with a bring-up overlay. Before a Pixel 7 DTB turns /trusty on by default, it should get the same kind of guarded start. Worth a line in the plan as the condition for that step.
  • The local HBM line is in place, but the panel driver on omamobile doesn't have that parameter yet. It comes with your panel commit that isn't proposed. Saying so in the plan or the README will save the next person some head-scratching.

On versions: our own #440 is omamobile-fingerprint 0.1.16 and goes first, so 0.1.17 here is right. device-panther still reads 0.1.9, which main ships; with #419 merging first at 0.1.10, this one becomes 0.1.11 and needs a rebase onto it then (README, install and the test; keep both sides). I'll write the squash message, since the first commit says 0.1.16.

Thanks also for adding the sg blacklist; it's good to have that on the Pixel 7 too.

derekross added a commit that referenced this pull request Nov 7, 2026
The matrix re-read against what merged and ran on hardware since 2026-11-02:
- Pixel 7: Bluetooth and the torch move to testing (kernel #196, #197, #426); fingerprint moves to WIP (kernel #207; the package side, #458, is in review); AVB footers and reboot=warm noted.
- Pixel 8a: a new row, from first light and UFS behind the guard (kernel #192, #206; #414).
- Pixel 6 Pro: GPU scaling and cooling (kernel #189) and UFS suspend (kernel #188) merged.
- Pixel 9a: GPU levels merged through kernel #96; GNSS's moving-car session (#114) and the link driver (kernel #185).
- Pixel 7a: Bluetooth's device tree merged (kernel #98).
The Pixel 8 Pro's row is left to #472.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@derekross

Copy link
Copy Markdown
Owner

@angelrenefm, on the guard question: yes, the runner may accept an empty persist:/ss/0 as the TDP copy on the Pixel 7. Stock has served that same empty file since the factory, so serving it unchanged is what stock does, any change on either side still stops the start, and nothing changes on the 7 Pro.

One condition, the test from my earlier comment: a case where stat reports a regular file with a size (say 4096) but dump produces nothing, which must refuse with no copy and no record, and one where stat reports something that isn't a regular file. That's the line between "empty" and "unreadable", and it's what keeps a failed read from handing Trusty an empty TDP while persist holds a real one. With that test and the status line updated, this is ready for its versions at merge.

The Pixel 7's vendor image ships the 7 Pro's fingerprint files under the
same names (g7.app, the Goodix HAL, the C++ Trusty KeyMint and
Gatekeeper, storageproxyd, trusty_apploader), and its stock dtbo has the
7 Pro's sensor node. What differs is the panel and the touchscreen.

The runner fell back to cheetah on any board but the 9a. Detect
google,gs201-panther and give it g7.app, the C++ KeyMint, its panel
(1080x2400 at 60 Hz, ABL's panel name sdc-s6e3fc3-p10) and its
touchscreen ("FocalTech FT3658"); the touchscreen name is now per phone.
props-panther.conf sets ro.hardware=panther. extract-fingerprint-vendor
takes -d panther with the 7 Pro's file list.

On the Pixel 7 with this package's files copied by hand (not pacman -U),
enrollment and unlock passed (device/panther/FINGERPRINT-PLAN.md, P5). test/all
passes, and fails when panther's touchscreen name is changed.

I used AI help (Claude) to work on this.
…ates

The Pixel 7's stock dtbo (board 0x30406) has the same goodixfp node,
reset pin and LDO15S supply as the 7 Pro's 0x30306, and its vendor image
the same trusted app and HALs. The plan records that evidence, what
differs (Trusty not yet started on panther, the S6E3FC3 panel without
local HBM in our driver, the FT3658 touchscreen), and the gates P1-P5
with their stop conditions.

The overlays enable Trusty, then the sub-PMIC and the sensor, on a
kernel with the panther fingerprint DT commit. Compiled with dtc and
applied with fdtoverlay: they change exactly the three status
properties, and the sensor overlay refuses a DTB without the commit.

I used AI help (Claude) to work on this.
The Pixel 7's S6E3FC3 panel gets local HBM in the kernel branch
panther-lhbm, behind the same allow_lhbm module parameter as the 7 Pro's
s6e3hc4. Add its line to omamobile-fingerprint-lhbm.conf; test/all
checks it as it does the other two, and fails when it is removed.

On the Pixel 7 the same line ran from a hand-written modprobe.d file,
not this package's: the panel lit the circle and enrollment passed (P5).

I used AI help (Claude) to work on this.
omamobile-fp's unit loads goodix_fp and its runner loads sg when it needs
the RPMB LUN; udev would otherwise load sg for every SCSI device (raw
SG_IO nodes past the read-only guard) and goodix_fp whenever the sensor
node is enabled. The lines and their reasons are the 7 Pro's.

Unlike the 7 Pro's file, Trusty is not blacklisted. There the sound
start (omamobile-aoc-start) loads it; this package has no such loader,
so with the trusty_* lines nothing would create /dev/trusty-ipc-dev0
and omamobile-fp would be skipped by its condition. Trusty loads from
the DT's /trusty node, as it did on the Pixel 7 where P5 passed.
test/all checks that both modules stay blacklisted and that no trusty
line is added, and fails either way.

On the Pixel 7 this file, copied by hand to /usr/lib/modprobe.d (not
pacman -U), went through a reboot: Trusty loaded from the DT,
omamobile-fp started by itself and the finger unlocked the phone (P5).

I used AI help (Claude) to work on this.
Trusty started on the Pixel 7 on 2026-11-03 under the sound bring-up's
overlay ("trusty version", "selected api version: 5", the trusty_ipc
vrings); what P2 still has to show is the device node and the cpu 7
line. Local HBM for the S6E3FC3 is now a local kernel commit, from
Google's GPL panel driver; section 4 says what it does and what to
check on the phone.

I used AI help (Claude) to work on this.
…O15S

The camera overlays enable the sub-PMIC too, so the base DTB keeps
LDO15S disabled, as it does the camera's LDO12S, and the sensor overlay
enables it. P1 to P4 passed on a Pixel 7 on 2026-11-04 (section 6):
Trusty answers, its keymaster, gatekeeper and apploader ports accept a
connection, and goodix_fp switches LDO15S on while the sensor is open
and off after, with no other rail changing. P5 is not run.

I used AI help (Claude) to work on this.
With the NDK path of fingerprint/Makefile (r27c), bionic declares
pipe2() only under _GNU_SOURCE, and -Werror stops the bridge on the
implicit declaration. Define it first, as the shim does. Both programs
then build without warnings, and test/all passes with them.

I used AI help (Claude) to work on this.
On a Pixel 7, persist:/ss/0 is a 0-byte regular file whose ctime is from
the factory (1 Jan 2020), in the 2026-11-03 and 11-04 backups alike:
its Trusty never committed TDP, and stock's storageproxyd serves that
empty file on every boot (init.gs201.rc links /data/vendor/ss/persist to
persist's /ss). The runner refused it, as it refuses a missing /ss/0.

Take an empty regular /ss/0 (debugfs stat: regular, Size 0) as an empty
TDP copy and record its hash; a missing or unreadable one is still
refused, and the existing checks still refuse once the copy or
persist:/ss/0 changes. test/preflight covers it, and fails with the old
runner.

Run on the Pixel 7 on 2026-11-04 and 11-06 (FINGERPRINT-PLAN P5): the
stack came up, a finger enrolled and unlocked, and the TDP copy was
still the empty file afterwards.

I used AI help (Claude) to work on this.
The preflight test now covers the line between an empty persist:/ss/0
and one that can't be read: a dump that yields nothing while stat says a
regular file of 4096 bytes, and a /ss/0 that isn't a regular file. Both
are refused with no copy and nothing recorded. Removing either the size
or the type check from persist_dump_tdp() now fails its case.

The Pixel 7's plan: the status line says P1 to P5 passed, the kernel is
linux-omarchy-mobile#207 instead of a local branch and commit, the
results are section 5 (there was no section 5), the panel driver on
omamobile doesn't have local HBM yet, and Trusty gets the 7 Pro's guarded
start before a Pixel 7 DTB turns it on by default.

I used AI help (Claude) to work on this.
@angelrenefm
angelrenefm force-pushed the panther-fingerprint-pr branch from b25eb10 to 2a090ab Compare November 7, 2026 15:00
@angelrenefm

Copy link
Copy Markdown
Contributor Author

Thanks, @derekross. Both are in, and the branch is rebased onto today's main (with #475 in it; no conflicts).

The test. test/preflight's fake debugfs can now fail a dump (it writes nothing and still exits 0, as debugfs does) and say what stat reports. Two new cases:

  • stat says a regular file of 4096 bytes, the dump yields nothing: refused, with no TDP copy, no provenance and no marker.
  • stat says Type: directory, Size: 0: refused the same way.

I removed each line from persist_dump_tdp() in turn: without the Size: 0 check the first case fails, without the Type: regular check the second one does. With both, omamobile-fingerprint's test/all passes (19), and so do omamobile-device-panther's (19) and tools/pr-check.

The plan.

  • The status line says P1 to P5 passed, and points at linux-omarchy-mobile#207 instead of my local branch and commit (in the gates section and the table too).
  • Results are now section 5.
  • Section 4 says the panel driver on omamobile has no local_hbm_mode yet; it comes with a panel commit I haven't proposed.
  • After the results: before a Pixel 7 DTB turns /trusty on by default, it gets the 7 Pro's guarded start.

I left the versions to you, as you said: omamobile-fingerprint still reads 0.1.17, and device-panther isn't touched (0.1.11 at merge).

I used AI help (Claude) to work on this.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @angelrenefm. Both tests are exactly what I asked for. I ran them myself with each check removed in turn: without the size check the 4096-byte case fails, without the type check the directory case fails, and both also confirm there's no copy and no record afterwards. That's the line between "empty" and "unreadable" pinned down, so the guard change is in with my yes. The plan reads right now too: the status line, linux-omarchy-mobile#207 in place of your local branch, the guarded start as the condition for turning Trusty on by default, and the note about local HBM.

I'll set the versions when I merge, as I said: omamobile-fingerprint 0.1.17 and device-panther 0.1.11, since main already ships 0.1.10 from your Wi-Fi PR. I'll write the squash message too.

Two small leftovers in the plan, no need to hold anything for them: the Panel row in section 2 still says local HBM isn't on the phone yet, and the end of section 4 still says to check the circle and the timeout "before P5". Whenever you next touch the plan. And when you get a chance after this lands, one start with the packages installed through pacman -U, rather than copied by hand, would close the last gap in the results.

derekross added a commit that referenced this pull request Nov 7, 2026
…he Pixel 7 (#458) (#485)

The Pixel 7 Pro's fingerprint stack (Trusty TA, HAL and the guarded runner) on the Pixel 7, with per-device values only where the Pixel 7 differs (panel, touchscreen). The runner accepts an empty persist:/ss/0 as the TDP copy, as stock has served it since the factory; tests pin the line between empty and unreadable (a non-empty file whose dump fails, and a non-regular file, are refused with no copy and no record). P1 to P5 ran on the contributor's Pixel 7: an enrolment that survives a reboot, with the unit starting the stack by itself. device-panther 0.1.11 because main ships 0.1.10.

Co-Authored-By: Ángel René <17342133+angelrenefm@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@derekross

Copy link
Copy Markdown
Owner

Merged through #485 as omamobile-fingerprint 0.1.17 and device-panther 0.1.11, with you credited as co-author. Thank you, @angelrenefm! Fingerprint on a second phone.

@derekross derekross closed this Nov 7, 2026
derekross pushed a commit that referenced this pull request Nov 8, 2026
The two leftovers from #458's review: the Panel row and the end of
section 4 still said local HBM wasn't on the phone. It has been since
2026-11-06 (kernel 7.3.0-rc6-omamobile-push1-g44048a9aa795): allow_lhbm
Y, local_hbm_max_timeout 3000, the HAL places the sensor at (540, 1713)
radius 100, the driver's centre, and the bridge's "LHBM on refused" came
only with the screen dark. The timeout itself is not checked.

I used AI help (Claude) to work on this.
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
The matrix re-read against what merged and ran on hardware since 2026-11-02:
- Pixel 7: Bluetooth and the torch move to testing (kernel derekross#196, derekross#197, derekross#426); fingerprint moves to WIP (kernel derekross#207; the package side, derekross#458, is in review); AVB footers and reboot=warm noted.
- Pixel 8a: a new row, from first light and UFS behind the guard (kernel derekross#192, derekross#206; derekross#414).
- Pixel 6 Pro: GPU scaling and cooling (kernel derekross#189) and UFS suspend (kernel derekross#188) merged.
- Pixel 9a: GPU levels merged through kernel derekross#96; GNSS's moving-car session (derekross#114) and the link driver (kernel derekross#185).
- Pixel 7a: Bluetooth's device tree merged (kernel derekross#98).
The Pixel 8 Pro's row is left to derekross#472.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
…he Pixel 7 (derekross#458) (derekross#485)

The Pixel 7 Pro's fingerprint stack (Trusty TA, HAL and the guarded runner) on the Pixel 7, with per-device values only where the Pixel 7 differs (panel, touchscreen). The runner accepts an empty persist:/ss/0 as the TDP copy, as stock has served it since the factory; tests pin the line between empty and unreadable (a non-empty file whose dump fails, and a non-regular file, are refused with no copy and no record). P1 to P5 ran on the contributor's Pixel 7: an enrolment that survives a reboot, with the unit starting the stack by itself. device-panther 0.1.11 because main ships 0.1.10.

Co-Authored-By: Ángel René <17342133+angelrenefm@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
…s#495)

The two leftovers from derekross#458's review: the Panel row and the end of
section 4 still said local HBM wasn't on the phone. It has been since
2026-11-06 (kernel 7.3.0-rc6-omamobile-push1-g44048a9aa795): allow_lhbm
Y, local_hbm_max_timeout 3000, the HAL places the sensor at (540, 1713)
radius 100, the driver's centre, and the bridge's "LHBM on refused" came
only with the screen dark. The timeout itself is not checked.

I used AI help (Claude) to work on this.
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.

2 participants