Alp SDK v0.11 — the Yocto image bakes end to end, and the gates that were green-by-skip now enforce
A full alp-image-edge bake for the e1m-v2n101-a55 machine now completes — 11168/11168 tasks — after three separate Yocto defects that had blocked every A-cluster build. Alongside it, a set of CI gates that were reporting green while enforcing nothing were made to actually run, and two AEN silicon-path bugs were fixed.
Alp SDK v0.11 is a correctness release. There are no new features in this cut: every entry in the changelog is a fix, a doc split, or a retirement. What it buys you is a Linux image that actually builds and a set of gates you can trust.
The headline is the Yocto side. A full alp-image-edge bake for the e1m-v2n101-a55 machine now completes — 11168 of 11168 tasks succeeded — after three defects that had each blocked it. Every A-cluster machine had been dying at base-files do_package with "tar: ./var/lib: Cannot mkdir: Bad address", 0 of 200 runs: GNU tar 1.35 creates its directory fd via the openat2 syscall, which the vendored pseudo 1.9.0 does not wrap, so pseudo forwarded a NULL path to the kernel and the kernel correctly rejected it. Getting past that exposed the alp-lvgl-dashboard recipe failing to find lv_conf.h and failing to link libdrm, and then an illegal libdrm RDEPENDS on the allarch packagegroup-alp-display. Each was fixed against a real rebuild of the failing task, not a CMake configure.
The second theme is gates that were passing without checking anything. The SB_CONFIG_* symbol gate — added to make invented sysbuild Kconfig names impossible — resolved its oracle from a Zephyr workspace and skipped when it found none; no job running it had one, so it shipped, skipped everywhere, and reported green. It now runs on the twister shard that already has the pinned Zephyr, and an environment flag turns a skip into a hard failure for any job that promised a workspace. In the same spirit, pr-bitbake now triggers on the example source its recipes actually compile (a Yocto-breaking edit had previously landed with no bake at all), and the ABI snapshot gate stopped regenerating the frozen v0.9 baseline against current headers.
Two fixes touch the AEN silicon path. xHCI DMA now translates M55-HP TCM addresses through the active core’s device-tree alias rather than a hardcoded HE window, and the Alif PDM FIR coefficient bank is written through sys_write32. Documentation gets the 1500-line board-config.md split into a landing page plus four references, and the superseded CC3501E integration plan is retired.
Status is unchanged and worth restating: v0.11 adds no new silicon-verified SoM family. Two families carry hardware evidence — the E1M-X V2N and the E1M-AEN801 (Alif Ensemble E8). i.MX 93, V2M/DEEPX, and the AEN30x–70x parts remain [UNTESTED] on real hardware. The completed bake is a build result, not a boot: it means the image compiles end to end for e1m-v2n101-a55, not that it has been flashed and run on the bench.
