M3-AUDIO-01 integrated candidate — offline result

Historical build result. AUDIO-01 was deployed and exposed a PCM notification bug. AUDIO-02 fixes it and now has clean44.1kHz physical playback. See the current audio result. The original artifact hashes and offline status below describe packaging time only.

2026-09-10. PASS offline; physical M3 audio remains untested. Owner manual BOOTIMG-only deployment is the next boundary. No device flash, reboot, raw codec or PMIC write, reconnect experiment or Y2PlayerNative implementation occurred.

Qualified DEV-02 baseline d76e57f; current source commit is the Git revision containing this result and the implementation. Final output result.json records that exact Git HEAD after commit; source-manifest.json hashes only the changed source delta against DEV-02. No complete donor/kernel/ROM hash pass was repeated.

Architecture/source review · Exact deployment and first commands · Retained validation · Physical entry evidence.

Output: /home/luca/Dokumente/Code/Y2Linux/out/y2linux-m3-audio-01/. Kernel 6.18.0-y2linux-m3audio01, Buildroot2025.02.17, ALSA utilities1.2.13.

ALSA card Y2Audio, PCM0 Y2 Headphone Playback, endpoint hw:CARD=Y2Audio,DEV=0; initially stereo S16_LE at44100/48000Hz. DL1->O00/O01-> CON3->upstream CS43131, VGP2 regulator1.8V, ordered GPIO20/18/15. SpeakerGPIO8 held low; headphone disabled until manually enabled. Master207=-24dB, generated three-second WAV=-30dBFS with fades. No speaker/jack/high-resolution acceptance.

Validation actually performed

One clean kernel tree and one fresh Buildroot tree; in-place fixes for Linux6.18 API differences and reviewed allocation/shutdown behavior. Final kernel compile has no warnings/errors. Historical failed compilation diagnostics remain in the complete compressed log. Buildroot completed, including ALSA/libdrm test utilities. A harmless missing documentation formatter in a host dependency was ignored by that upstream build; it did not affect installed artifacts.

All 13 targeted tests passed, covering production trigger/IRQ fault injection, 32-bit DMA/ring boundaries, both rate codes, CCF preservation of adjacent SD fields, actual PWRAP masks/failures, resolved audio DT mutations, preserved RAM and eMMC protections, rescue isolation, I2C combined transfers and ARM ABI. Cortex-A7 QEMU runs pass the ABI check and aplay/amixer version execution. Shell scripts pass target-BusyBox syntax checking. No software test claims real DMA timing, I2S pads or analog sound.

Emitted ELF/Image/zImage/module/DT/rescue and BOOTIMG checks pass. Linux bank and all reservations match DEV-02. Before kernel reservations, allocator-described RAM remains997179392bytes (951MiB minus16KiB). Resident kernel span grows from 10626240 to10910080bytes, about277KiB; next-boot actual MemTotal must be recorded, not forced to equal DEV-02's954660KiB. Four CPUs/HIGHMEM/core config stay enabled. BOOTIMG is below the16MiB partition limit, including validated LK read padding.

Both rootfs tar and512MiB ext4 contents pass checks; read-only e2fsck passes. ELF ABI, key permissions/policy, matching module, aplay/amixer/speaker-test/evtest, fixture formats/duration/peaks and no mixer auto-restore are checked. SSH host keys are not generated into the image; the owner's authorized public key is retained. Existing removable SD content needs an offline update from rootfs.tar, preserving its persistent SSH keys. No partitioning/formatting is required.

Boundary audit and actual physical status

Rechecked current coverage against DEV-02 physical evidence and explicit owner M3 authorization. Core SD/SSH/initial USB, four CPUs/expanded RAM, visible display, wheel/buttons and PMIC/core buses retain their narrow physical acceptance. M3 source/build readiness advances, physical audio does not. #29 stays open and active. #27 reconnect remains explicitly deferred under #28 for later whole- platform qualification. #30 full power, #31 connectivity and #32 GPU/final platform work precede Y2PlayerNative. No older research task reopens qualified core acceptance. After owner deployment, classify A–G failures by layer and require clean audible output plus stable DMA/IRQ/CON3/power and repeated start/stop/rates.