Research / incomplete — this repository has produced no working Linux fingerprint enrollment solution. No CV2-004 exists. Decision B remains. The original owner is no longer actively pursuing the project. This is a handoff for experienced protocol/hardware reverse engineers and agents working under their supervision, not a driver release or installation guide.
Target: Dell Latitude 5495, Broadcom BCM5880 / Dell ControlVault 2,
USB 0a5c:5834. Windows Hello historically worked on this physical reader.
The machine is now Arch-only; that Windows installation is gone.
Most reverse engineering, binary analysis, experiment design, implementation and technical documentation was performed agentically by coding/reasoning agents under the repository owner's direction. The owner is a software developer and does not claim expertise in ControlVault internals, cryptography, hardware reverse engineering or the low-level conclusions presented here.
This is evidence provenance, not expert certification. Conclusions may contain mistakes. Independently verify addresses, disassembly, protocol structures and control flow before relying on them. Experimental patches are not trustworthy merely because their behavior was documented or mocked tests passed.
We preserve these labels:
- FACT: observed in a reported hardware run or directly supported by a specified static artifact. Static FACT labels are agent findings, not independent expert verification. The evidence source must be stated.
- INFERENCE: interpretation supported by evidence, not directly established.
- HYPOTHESIS: requires further testing. Never silently promote it to FACT.
FACT — reported Linux hardware tests: discovery, communication, fingerprint
capture and multi-stage enrollment progression worked with compatibility patches.
Finalization reached command 0x6e, returned 0x8d (decimal141), and left
templateHandle = 0. No usable enrolled template was established.
Accepting 0x8d at success/decoder gates produced apparent enrollment completion
with handle zero and verification failure. Status remapping is not a fix.
A local fprint record or enroll-completed does not prove a device template.
INFERENCE: the failure is in generation-specific protocol/session support, not merely fprintd, PAM, KDE or the UI. The upper stack reached the failing operation, while later experiments exposed an unimplemented local SDK function. Changing desktop authentication configuration cannot supply that missing code.
FACT — agent-derived static analysis of the pinned Dell A21 package: the
Windows BCM5880 branch captures features through 0x5c, combines four feature sets through 0x6f, creates a
ControlVault type-7 object through 0x09, receives a four-byte handle, and stores
a reference on the host for later matching. The ordinary 0x6e path is bypassed
in the analyzed branch. Linux cv_create_object is a logging stub returning
0x73. These findings describe analyzed code, not a newly captured Windows run.
HYPOTHESIS: reproducing this workflow with the required sessions could produce genuine Linux enrollment. It has not succeeded on hardware.
| Experiment | What it established | What it did not establish |
|---|---|---|
| CV2-001 | Isolated host-managed candidate passed mocked/offline checks. Hardware run still failed; selector mismatch strongly explains fallback to the legacy worker | Successful new capture/finalization or valid template |
| CV2-002 | Corrected exact-field selector worked on hardware. Anonymous capture open failed locally with 0x00080006 |
No 0x5c, 0x6f or 0x09 through the new path |
| CV2-003 | Instrumentation-only candidate; offline tests reproduce the local secure-setup failure | No recorded new hardware success; not a fix |
FACT — agent-derived static analysis: Windows setup uses
0x23 → 0x24 → protected 0x02 before temporary capture. Linux
cvhSetUpSecureSession returns 0x00080006 locally. Host command protection
and response unprotection are also incomplete. A setup-only patch is insufficient.
Decision B: no justified complete secure-channel candidate. Missing evidence:
- Actual runtime-selected legitimate public trust material and verified branch.
- Independent Windows-origin crypto/channel inputs and outputs, including
protected
0x02and authenticated response behavior. - Safe device-side recovery after a partially completed handshake.
Offline crypto/model tests passed, and OpenSSL reproduced a package-resident AES vector. Neither proves Windows channel interoperability. A bounded package scan decoded21 certificates, all RSA; it did not establish the runtime DSA certificate. Traced host cleanup does not establish device-state reclamation. No inaccessible private-secret dependency has been demonstrated.
Run a small public-input native Windows crypto API harness on an already available x64 Windows computer, or obtain equivalent provenance-complete Windows results. No Dell driver, reader or local ISO is needed for that limited API stage. The collector is specified but not implemented or Windows-tested. See the self-contained collector specification. A local Windows VM remains an option, but the owner's ISO/download constraint has paused it. Wine, emulator stubs and synthetic roundtrips are not validated substitutes for Microsoft-runtime behavior.
Later, a fresh working A21 installation on the actual reader could supply new trust/channel evidence; reconstructing the deleted historical installation is not required. Even a successful normal open/close would not prove partial-state recovery. No further hardware experiment is authorized by this release.
- Continuing the research — state, safety and agent prompt.
- Technical handoff — address anchors, hypotheses and gates.
- Experiment summary — reported outcomes and limitations.
- Local artifacts and reproduction — lawful sourcing and hashes.
- Licensing and publication scope.
The curated public export includes the MIT license for repository-authored material within the owner's legitimate licensing scope. It does not claim ownership of or grant rights to Dell, Broadcom, Canonical, Microsoft or other third-party/vendor material. Excluded binaries, firmware, extracted resources, disassembly and raw captures are not covered merely because this research discusses them; see LICENSE-NOTES.md.
This is a curated source/documentation handoff. Vendor binaries, firmware, patched libraries, full disassembly and private captures are deliberately absent. The old research history is not the public history; hashes/checkpoint IDs identify provenance, not files or commits guaranteed to exist in this export. Original evidence must be regenerated from independently obtained local inputs.
Do not install the historical candidates, fake success, suppress signature/MAC checks, alter fingerprint verification, or send guessed recovery commands. Never restart system D-Bus: a prior experiment disrupted the graphical session. No sudo/root, firmware changes, live library replacement, service changes or physical interaction without a separate explicit hardware plan and approval.
Success requires genuine device-side enrollment, a usable identifier (normally a nonzero handle unless an alternative is proved), repeated actual fingerprint matches, and persistence across fprintd restart, preferably reboot. Metadata, synthetic tests and optimistic status handling are not substitutes.
No success is promised and there is no basis to say it is imminent.