03-rstar000-synaTudor-progress.md 3.0 KB

Draft: rstar000/synaTudor — 06cb:0081 progress + capture-ioctl crash forensics

Where: https://github.com/rstar000/synaTudor (upstream: Popax21/synaTudor — feel free to forward) Type: issue (progress report + question; patch set attached) Attach: work/patches/synaTudor-work.patch (vs commit 994c6ba)


Title

06cb:0081 (Lenovo YOGA 720-13IKB): init, command loop and datastore work — capture ioctl crashes on a NULL dispatch-map entry

Body

I picked up this project for a 06cb:0081 (Validity-family "Tudor" sensor, Lenovo YOGA 720-13IKB, WBDI driver 5.5.2731.1050) and got substantially further than the previous state, but I'm now stuck in the driver's capture path on something I suspect the host must populate. Sharing in case it helps anyone else — and happy to hear ideas.

What works now (patch set attached, 12 files):

  • Driver init: Attach → Reset → QueryStatus (status 0x5 = NOT_CALIBRATED until calibration is triggered)
  • Command loop + datastore round-trip (tudor_cli reaches its prompt)
  • USB protocol, TLS handshake, firmware handshake all functional

Main fixes: accurate WINBIO_SENSOR_INTERFACE_V1 table layout (15 slots: Attach, Detach, ClearContext, QueryStatus, Reset, SetMode, SetIndicatorStatus, GetIndicatorStatus, StartCapture, FinishCapture, ExportSensorData, Cancel, PushDataToEngine, ControlUnit, ResetOwnership — no PipelineInit/Activate), whole-table restore from pristine RVAs, well-formed empty CalibrationData PROPVARIANT, re-acquire across USB re-enumeration, correct tudor_devctrl_wudf1 completion semantics, ~30 output-pointer fixes in the WUDF methods, AssignContext/RetrieveContext, live factory objects.

The blocker — capture ioctl 0x44202c crashes at driver offset 0x34314 (call chain 0x36b00 → 0x36cd9 → 0x34360 → 0x3414c) dereferencing a NULL at [ctx+0x1c0].

Forensics:

  • [ctx+0x1a0..0x1e8] looks like an MSVC container layout ({data-ptr, …, count=2, 0x30}); [ctx+0x1c0] is element 4/notional index and the driver indexes into it for the capture entry, then dereferences the element's vtable (+0x38).
  • The three writers of that field I could identify (0x199d2, 0x1cc61, 0x6ea49) never execute under my host.
  • Host vtables match the official UMDF 1.1x wudfddi.h interfaces, and the driver only implements IPnpCallbackHardware + IPnpCallback.

Hypothesis: this is a dispatch/handler map built during device-interface setup (CreateDeviceInterface / AssignDeviceInterfaceState / io-queue creation), and my host's no-op interface handling leaves the capture entry NULL. Guidance on what the UMDF host does at interface activation that fills this map would be very welcome.

Workaround, for context: the same sensor fully works under uunicorn/rstar000's Wine-sandbox approach (vendor Windows driver + a hacked Wine 4.8 with a fake libusb-backed winusb): full enroll + verify, currently wired into PAM for sudo. A native solution would still be much better (no Wine), so I'd love to see this unlocked.