[chriskearney]

Release pins for aurora-2026.10.17.20. The kernel is c46d7dcd, the current integration/aurora tip (.19 plus #136), built with the release 19 recipe and config. Native OpenGL keeps source d616d065. Its package release 10 differs from release 9 only in the session check, which admits this kernel's marker.

Changes

• assemble-m3-stack.py: the Pro performance kernel and its marker now name c46d7dcd. The Neo and J615 allow-lists gain it, as release 19 added its own.
• pro/mesa-m3-pro-check: admits exactly the release 16, 18, 19 and 20 markers. The native and Vulkan sources and the T6030 identity check are unchanged.
• recipes/release20: the kernel, native and Vulkan recipes. MESA-SOURCES.md points at them.

Tests

• Kernel package (b1017-114018-705d): 25 checks pass.
  • The recipe config is release 19's. olddefconfig changes 16 symbols: user4k is compiled in (off unless booted with user4k=on), the separate Neo graphics symbols are gone, and DRM_GEM_DMA_HELPER becomes a module.
  • The modules are release 19's plus drm_dma_helper.ko.
  • Only the J700, J813 and J873G device trees changed.
  • The prepared config, file list and all 115 device trees are identical to the lab package of 7638540a that booted on the M1 Air and M3 Pro.
• Tool tests: test_m3_pro_perf, test_m3_stack_assembly, test_release and test_neo_gpu pass (108). The full suite runs 917 tests with 5 skipped. It fails only the 5 test_t8112_sep_dma errors that come from the host dtc; with the kernel's own dtc (DTC=) those pass 3/3.
• Neo Mesa: mesa-neo release 9 carries this tree's mesa-neo-env and drops the render-node udev rule. Its Mesa source is unchanged, so on this kernel the Neo uses software rendering until a Neo Mesa built for the new GPU interface ships.
• Generated installer: 274 tests: 271 pass, 3 skipped. 22 assets, and all 16 package pins match.
• M1: no M1 lab Mac was available for the .20 package. The lab package of the identical tree (7638540a, b1017-102213-c4bd) had 5 clean boots on an M1 Air, 3 of them on a TS4 dock and 2 with user4k=on, with omarchy-m-test showing no regressions.
• M3 Pro (j1017-121031-7fe6, with native package 10, on the .20 package): all 85 native pixel cases pass, along with GL/Vulkan readback, the native compositor with two CRTCs (eDP + DP), the display coprocessor gate (same as .19), and the session/performance checks. The vkQuake timedemo runs 969 frames at 219.4 fps. omarchy-m-test shows no regressions, and there were no AOP or GPU faults.

🤖 Generated with Claude Code