Skip to content

asahi: admit the measured 30-core T6034 G15C contract - #32

Closed
chriskearney wants to merge 1 commit into
omacom:integration/aurorafrom
chriskearney:chris/j514m-t6034
Closed

chriskearney wants to merge 1 commit into
omacom:integration/aurorafrom
chriskearney:chris/j514m-t6034

Conversation

@chriskearney

@chriskearney chriskearney commented Nov 5, 2026 •

Copy link
Copy Markdown

J514M's binned M3 Max reports chip 0x6034, so the existing 0x6031-only G15C check rejects it. Admit its measured contract: B1 revision, one die, four physical ten-slot clusters, USC3, legacy HAL, nonempty ten-bit core masks totaling 30 enabled cores. The fresh powered kernel-loader probe found cluster counts [8, 8, 7, 7]. Existing T6031 admission and native kernel ABI checks remain unchanged. Startup requires no environment or enable flag.

The change is based on installed production source a19691114759e2c0507f3502cf6166ef2c6e2e33, with only this admission change and its tests. Candidate source: 2c893e7abed3bd91f9a06821a549590fde5a23b7.

Validation:

  • Actual chip-admission host suite passes, including wrong chip/revision, core slots, clusters, dies, ISA, HAL, empty/out-of-range masks and 29/31-core negative controls.
  • Isolated ARM64 production split-package build b1105-160343-6ea5 passes all 129 configured Meson tests, including asahi-chip-admission. Candidate version: 1:26.3.0_devel.20261105.231484.2c893e7abed-1.60341.
  • The normal /usr recipe's LLVM/SPIRV checks were required explicitly despite the builder's --nocheck default. Archive metadata and direct ELF inspection confirm the original runtime dependency set, exactly one libLLVM.so.23.1 in libgallium, and no LLVM/SPIRV-Tools runtime dependency in Honeykrisp.

Native rendering and boot qualification on J514M remain pending; this draft does not claim hardware correctness or channel promotion. No device packages were installed by this source/package build.

The J514M T6034 reports G15C with four physical ten-slot clusters and 30 fused cores. Admit that exact chip and B1 revision with nonempty ten-bit masks totaling 30 cores, one die, USC3 and the legacy HAL. Preserve the existing T6031 path and all native kernel ABI requirements.

The chip-admission host test checks the measured masks and rejects wrong chip, revision, slots, clusters, dies, ISA, HAL, empty/out-of-range masks and 29/31-core layouts. This is source and host admission evidence; hardware correctness still requires the candidate package on the affected Mac.

Generated-by: LLM

Assisted-by: Codex:GPT-6
@BTNG-studio

Copy link
Copy Markdown

Tested on a second T6034 machine: the 16-inch J516M (MacBook Pro 16", binned M3 Max, 30-core GPU). This PR's admission change works there unchanged; no J516M-specific Mesa change is needed.

  • Mesa: this PR's agx_chip.c / agx_device.c change applied as a patch on a19691114759 (package mesa-aurora 26.3.0_devel.20261031-101, tests not included in that local build).
  • Kernel: the J516M admission in J516M: drive the 16" M3 Max MacBook Pro's display and GPU by default (stacked on #372) linux-aurora#405, stacked on Add guarded J514M/T6034 native graphics bring-up linux-aurora#372. The GPU reports the contract this PR requires:
    asahi-g15 40a400000.gpu: GPU firmware build: J516M RTKit 2419.140.12 (local port, J514M identity)
    asahi-g15 40a400000.gpu: T6034 G15C: power acknowledged, version=0x7041100 counts=0x21040a core-mask=0x7b5ed7f5fd
    asahi-g15 40a400000.gpu: T6034 G15C: cores enabled per cluster: [8, 8, 7, 7]
    
    That's four ten-slot clusters with 30 enabled cores. (The local build's firmware row name was shortened for #405.)
  • Result: pass. Hyprland 0.56.2 renders on renderD128 (asahi-g15) with a high-priority context, on the internal panel at 3456x2234@120 and on a Studio Display at 5120x2880@60. In use across several boots since 2026-11-06, with no GPU faults in the kernel log.
  • Not tested: glxinfo/vulkaninfo (not installed), CTS, sustained load.

🤖 Generated with Claude Code

@maralcbr

maralcbr commented Nov 7, 2026

Copy link
Copy Markdown
Member

Carried unchanged as #33, now merged.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants