> **Historical record.** The dates, candidate identity, "current" claims, > next steps and permissions below belong to this recorded boundary. See > [current state](../../CURRENT_PLATFORM_STATE.md) for the latest physically observed result.

Current session checkpoint — 2026-09-10

Initial ACM/Ethernet/SSH work, including return after owner startup/restart. USB cable disconnect/reconnect still fails without restart. This issue remains OPEN and deferred to later platform qualification, not an M3 blocker.

September 10 session audit. Earlier phase ordering below is historical where superseded.

DEV-02 live hardware checkpoint — 2026-09-10

OPEN: physical USB reconnect failed; core/Buildroot otherwise substantially qualified.

DEV-02 live qualification retains baseline, completed 256 MiB test, input/display/SD and USB failure evidence. No reboot, kernel build or BOOTIMG was prepared. This checkpoint supersedes stale untested/blocked statements below within its measured scope.

Y2B-255 — Add bounded USB CDC ACM kernel and PID1 logging

Offline integration checkpoint — 2026-09-10

The proven ACM path and offline one-reconnect lifecycle are integrated into the combined baseline; shared PMIC/clock ownership and early USB startup replace competing mappings. New provider integration and reconnect remain unqualified on hardware.

M2-BASELINE-01: exact implementation, RAM decision, checks and owner test. The historical scope/sequence below is superseded where it conflicts with the owner’s combined integration instruction. All remaining acceptance criteria stay open.

Additional evidence — 2026-09-10

USBACM-03 remains our confirmed 11,754-byte kernel/PID1 capture. USBACM-04 and M2-INPUT-01 remain pending hardware results. The external project confirms ACM+ECM feasibility and documents the mc IRQ name and getty HUPCL/DTR pitfall, but supplies no equivalent bounded reconnect qualification. Our reader already clears HUPCL during capture. Preserve canonical guarded PIO MUSB/PHY and existing robustness tests; no donor PHY replacement, ECM expansion or new USB hardware claim.

Research matrix and source limits. M1 complete; M2 active/exit incomplete. External results do not promote our hardware status. Existing acceptance criteria and historical checkpoints follow.

Current execution checkpoint — 2026-09-09

The owner explicitly requested one combined enumeration/logging candidate to reduce physical iteration. Scope audit keeps M1 complete, M2 active/exit blocked and later phases deferred. CHRDET evidence establishes charger presence under the reported cable conditions; connected-start6a=BE still refuses wake. No new hardware result has been inferred.

Immediate next test: M2-USBACM-01, 1,159,168 bytes, SHA-256 d2d1f8cbbaa23ffcbdeddb317db7a8e62bc21deca9ebebc564c9b55a5590aedd. Reviewed combined contract: boot unplugged, perform the proven guarded wake, attach at the screen prompt, check fresh supply/clock/PHY/DMA state, apply the FM/LK digital device-session sequence, then attempt upstream MUSB peripheral PIO, one standard g_serial ACM and bounded retained/live kernel + PID1 logging. No host VBUS supply, analog calibration, PMIC writes, active-DMA recovery, RAM expansion or other partition writes.

One clean build passes 23 full-suite methods, one additional linked-syscall mutation method and 8 observation methods, including 30 QEMU ARM PID1 scenarios and host/ARM relay faults. Exact DT/config, linked accesses, D08 and BOOTIMG pass. Host capture tool is ready. No device operation was performed. Stop for the owner's BOOTIMG-only physical test within 60s. USBGUARD-01 remains an optional untested fallback, not a prerequisite test.

This advances offline implementation only. Physical enumeration, real log continuity, late-open/non-reading behavior and controller reconnect qualification remain outstanding. The first trial stops on first detach/error or 50s active window; reconnect and connected-start recovery are not implemented. #23/#27/#28 stay open, and every original hardware acceptance criterion remains intact. Original drafting-only restrictions below are superseded only for this specifically authorized/audited experiment, not a general production port.

Status, type and goal

Draft implementation contract; BLOCKED on the MT6582 USB controller gate below. Not implementation-ready for hardware, and not state:luna-ready. The upper-layer logging design is specified; no implementer is asked to choose PHY registers or invent a DT compatible. This issue does not absorb an unresearched controller port.

One goal: over the existing Y2 USB data port, retrieve retained kernel messages and continuing PID1 logs on a Linux PC, independent of screen reading or physical UART. Exactly one CDC ACM port; no interactive shell.

Research, source evidence and detailed contract.

Why now / dependencies

M1 hardware execution and periodic PID1 diagnostics are owner-observed (#20/#21). Host log access would shorten every later RAM, input, display and storage cycle. Y2E-160 #23's PMIC/supply research is a prerequisite source for VUSB, not a claim that USB power is already understood. Physical UART #16, full RAM expansion #22, input/display/SD implementation are not prerequisites. D08 remains the default. No reordering/execution of the five queued foundation issues is authorized here.

Hard gate before any runtime implementation: record an exact, reviewed MT6582 USB device-controller contract in docs/knowledge/usb-logging.md:

Controller MMIO span, SPI/polarity, interrupt handling and sufficient FIFO/
endpoint layout for EP0 + ACM bulk IN/OUT + notification IN.
Actual clock-provider/gate/48 MHz mapping, USB2 PHY register/initialization/
calibration contract and VUSB supply/ownership after normal LK handoff.
Peripheral/VBUS detection and safe disconnect/reconnect; no host VBUS sourcing.
Quiescence of inherited USB DMA before PIO, without touching display DMA.
Exact v6.18 glue/PHY compatibility or a reviewed completed prerequisite change.

Stock musb-hdrc.0 IRQ 64 and public MT6582 SPI32 evidence support the route. They do not prove that existing mediatek,mtk-musb glue plus a sibling PHY works. Upstream glue needs three clocks and a generic PHY; current Y2 DT provides neither. Do not unblock until this gate is resolved. If it needs a controller/clock/PHY implementation, stop and return the single smallest prerequisite for owner review; do not create further issues or expand this issue automatically.

Exact implementation scope after the gate

Reuse proven MUSB in peripheral-only PIO mode with built-in g_serial,
defaults ACM on / OBEX off / one port. Kernel intent: USB_SUPPORT=y, USB_GADGET=y, USB_MUSB_HDRC=y, USB_MUSB_GADGET=y, MUSB_PIO_ONLY=y, USB_G_SERIAL=y; host USB=n, DMA/OTG/dual-role off. USB_MUSB_MEDIATEK and NOP_USB_XCEIV apply only if the exact glue is verified; selected GENERIC_PHY/USB_ROLE_SWITCH and the real provider must be accounted for. U_SERIAL_CONSOLE=n; no console=ttyGS0 dependency. Existing PRINTK/TTY/ proc/sysfs/time32 stay enabled. g_serial selects LIBCOMPOSITE/CONFIGFS_FS and serial functions; no configfs mount or setup daemon is needed.
DT: only the reviewed USB/PHY/clock/supply nodes, dr_mode="peripheral".
Extend the exact DT validator whitelist, not D08 bounds. No guessed addresses, dummy PHY, fake fixed clocks, unrelated clock gating or full-SoC DTS import.
Existing native PID1: bounded raw nonblocking /dev/ttyGS0 relay. Discover
its dynamic device number from sysfs; read kernel ring through /dev/kmsg (1:11, read-only). Keep an 8 KiB PID1 startup/status ring, 32 KiB TX queue, 8 KiB kernel-record buffer. Service at most 16 records / 16 KiB each heartbeat pass. Preserve partial writes and report drops/gaps; boot/heartbeat never waits for a host. New syscall numbers require actual linked-kernel checks.
Host sends only LOG1\n to begin/restart replay of retained kernel and PID1
startup records, then live records/status. Bounded 16-byte parser; no shell, arbitrary command execution or writable filesystem/log API. Include kernel release, build identifier, source/sequence and explicit overflow markers.
Host script selects an explicit verified USB tty path, commonly /dev/ttyACM0
but never inferred from numbering alone. Raw 115200 8N1/no flow control, ACM DTR asserted, sends LOG1, records bytes/timestamps/USB identity and exit reason. Use upstream standard g_serial descriptors, not stock Android VID/PID. Read /sys/class/tty USB parent identity; no privileged global host-rule changes.

Files/components

Prospective edits only: kernel/config/first-boot.config, kernel/dts/innioasis-y2-first-boot.dts, tools/validation/dtb.py, initramfs/init.c and a small bounded relay helper; tools/observation/usb-log-capture.py, tests/test_usb_logging.py and affected PID1/syscall/artifact tests; docs/build/usb-log-capture.md, docs/knowledge/usb-logging.md. Reuse upstream MUSB/gadget code where the gate establishes compatibility. A broad USB/PHY/clock/PMIC driver port is a prerequisite, not hidden scope here.

Explicit out-of-scope

Implementation/build/flash during this drafting session; automatic hardware trials; ADB/adbd, networking, RNDIS/ECM, mass storage, Debian, storage access, app work, USB host/OTG switching, charging-policy changes, accessory UART muxing, preloader/ LK/security/partition changes, RAM expansion, full display driver, persistent ramoops reservations and guaranteed pre-enumeration crash/panic capture.

Acceptance criteria

Hardware prerequisite contract is reviewed and complete; no guessed driver,
compatible, clock, PHY/power or DMA handoff assumption remains.
Resolved config is peripheral/PIO/one-ACM, with expected dependencies and
real syscall implementations; only approved DT nodes are added.
Linux host enumerates the selected Y2 as CDC ACM and opens its tty without
custom kernel drivers; capture identifies actual USB path/descriptors.
One LOG1 capture includes retained early kernel-ring messages, PID1
startup/mount/sleep status and at least ten increasing heartbeat records. These are actual kernel logs, not only rendered-screen summaries.
Late host open, non-reading host and one authorized disconnect/reconnect
do not stall PID1; reconnection can replay and reports any overwritten logs.
Positive hardware results name candidate SHA-256, host tool/version, raw
log/hash and timing. Offline fixture success is not enumeration success.
D08, CPU0, watchdog and guarded framebuffer fallback remain intact;
BOOTIMG bounds pass and loader/partition/calibration state is untouched.

Do not close as hardware-success based only on source review or fixture tests. The issue may remain blocked while upper-layer intent is fully specified.

Validation

No implementation or device tests in this session. Later targeted host/ARM fixtures must cover partial/EAGAIN/EINTR/ENODEV/EPIPE writes, ring wrap/overrun, late LOG1, malformed/overlength input, partial lines, no-reader/reconnect and continued heartbeat. Check actual linked syscall targets, dynamic tty discovery and bounded memory/CPU budgets. Replay pre-enumeration kernel records if still retained; report gaps instead of silently promising lossless delivery.

Initial USB is a new hardware subsystem: review controller/PHY/clock/power and DMA safety, binding/config resolution, affected regressions and complete D08/layout/BOOTIMG validation under the broader verification boundary. Trust unchanged pinned source/toolchain/recovery provenance. Subsequent relay-only changes use one clean build, targeted tests and final D08/package size/hash, without routine historical full-suite or two-build repetition.

Any later device validation requires a separately authorized BOOTIMG-only trial, retained known-good fallback and bounded observation window. No flash is implicitly authorized by implementing or closing an offline checkpoint.

Stop conditions and alternative

Stop at the controller gate, unexpected PHY/PMIC/clock/reset dependencies, inherited-DMA uncertainty beyond the approved experiment, IRQ storm, VBUS-source risk, framebuffer/heartbeat regression, parser bounds failure or D08 overflow. Do not enlarge RAM or port an entire subsystem to make this issue fit.

No simpler reliable host-readable alternative is established: ADB, USB Ethernet and vendor bulk all require the same missing UDC/PHY; UART-over-USB requires an unverified electrical accessory, and ramoops lacks a proven retained range and reader. Preserve the working screen temporarily; do not call it USB logging.

Unlocks

Host-retained boot/runtime logs for faster single-change RAM/input/display/storage bring-up. It does not observe preloader/LK or hangs before USB/PID1 initialization, and it does not start another platform milestone or Y2PlayerNative work.