Skip to content

J614c (MacBook Pro 14-inch M4 Max) first boot: works, internal display on simpledrm, no GPU on a 40-core chip with 26.6.2 #1332

Closed

Description

@joske

I installed Aurora ALARM on a MacBook Pro 14-inch M4 Max (J614c, Mac16,6, 16-core CPU, so the 40-core GPU), using Aurora Installer 0.1.0 alpha 18. System firmware is 27.0 and the stub is 26.6.2. As far as I know this is the first J614c boot. It now boots from the internal SSD and is usable, with the two gaps below.

What it took to boot

With those, it takes about 15 s from the chime to GDM.

What works

The t6041-j614c.dts from #1140 boots unchanged. The IPSW device trees match its approach: J614c differs from J616c in exactly the 23 properties where J614s differs from J616s, and has the J614s value in each. That includes Wi-Fi module-instance = mriya.

Area Status
CPUs All 16 cores come up; apple-cpufreq with schedutil (cpu0 reports a maximum of 4104 MHz)
Storage NVMe works
Wi-Fi BCM4388 works, with "using bidirectional receive DMA"
Bluetooth Works
Keyboard, trackpad Work; tpmtfw-j614c.bin was found
Audio The headphone card is there; the speakers are off by design

1. Internal display: simpledrm only

aura's T6041 display handoff (kboot_display_handoff.c) only allows apple,j616c. With apple,j614c added, it gets further and then stops here:

DCP: T6041: display power state (ADT /arm-io/dcp): DISP_CPU 0x502290000 = 0x1f0003ff, on
DCP: T6041: CPU_CONTROL 0x10 (running) STATUS 0xec; PROTECT dart-dcp 0x1ff dart-disp0 0x1f9
DCP: T6041: display is not status "disabled"
DCP: T6041: armed, but handoff NOT published; display nodes untouched

The two sides disagree about the M4 Max:

  • aura's handoff expects /display-subsystem to be disabled, as on the J616s.
  • t6041-j614c-j616c.dtsi sets &display { status = "okay"; }. The kernel's M4 Max gate (gate_t6041_max_run) "never touches the internal display" and only enables the external chains.

So I think the J616c's internal display is also on simpledrm in the current integration tree. Is that intended for now, or is the DCP path for the M4 Max internal display somewhere I didn't look? I have the one-line aura allowlist change ready for when the J616c path is settled.

2. GPU: refused, so software rendering

aurora 28a600000.gpu: unsupported GPU firmware build: TEXT 0x60000 DATA 0x138000 ...
aurora 28a600000.gpu: supported: Apple GFX 25F84 for T6041 (normalized loaded image)
aurora 28a600000.gpu: Unsupported GPU firmware image (reported macOS version Some([26, 6, 2]))

This chip has the 40-core GPU and runs 25G83 firmware. The integration tree's T6041 profile is 25F84 on 40 cores. #1050's 25G83 profile was captured on a 32-core J616c and checks the core count and content hashes, so as I understand it, it won't start this GPU either. GNOME on llvmpipe at 3024×1890 is sluggish.

I'm happy to test builds or capture whatever a 40-core 25G83 profile needs from this machine. Tell me what to run.

Logs

The dmesg, journal and aura kboot logs are available. I can attach them.

🤖 Generated with Claude Code

Activity

  1. joske commented on Nov 9, 2026

    @joske
    ContributorAuthor

    GPU update for the J614c, which has a 40-core GPU (16-core CPU) and a 26.6.2 stub.

    The firmware image is different from the J616c's. On 26.6.2 this machine loads a T6041 GPU image with the same entropy offsets as the 25G83 entry in #1050 (0x59839/0x5991c), but a different digest:

    sha256 7c04360e05a6b8628f61262d0a244166187ad87a734716d8910908e94ffec08a
    

    #1050 has aa113ed6…, so I assume Apple ships separate 32-core and 40-core builds.

    What I tried:

    • On top of drm/asahi: add the T6041 25G83 profile captured on the J616C #1050 (rebased onto b1c9943, linux-aurora -28), I added a third T6041 build entry with this digest, the V25G83 variant and config: None. That uses the static 40-core configuration the 25F84 entry runs with.
    • mesa-aurora from aurora-wip (4fc7a6856c). The packaged ALARM Mesa (c1753704ed7) predates mesa#85 and refuses the M4 Max ("G16C uses the HAL200 command stream, which this driver does not implement").

    Result:

    • The firmware is admitted and RTKit comes up: GPU firmware: macOS 26.6.2 (25G83 …), 40-core, then [drm] Initialized aurora.
    • Vulkan enumerates Apple M4 Max (G16C B1).
    • But the first render job times out, and everything after it queues behind that job until reboot:
    aurora: GPU work of a client failed (ETIMEDOUT, reported at g17/runtime.rs:1316, client gnome-shell[861]); the client sees a lost device
    aurora 28a600000.gpu: [File 3 Queue 2]: IOCTL: submit failed! (submission ID: 130 err: EIO)
    aurora 28a600000.gpu: Render watchdog owner=9 vm=4 publication=33 renewals=0 decision=Defer behind earlier work of another VM: blocker_vm=2 blocker_publication=1 blocker_failed=1
    

    So the static 25F84 contents don't work with 25G83 on this chip. The GPU is blacklisted on the J614c for now.

    @AxxzyWasTaken, t6041_captured.rs says its contents come from the j616c/b25 driver's encoders, evaluated from the device tree, and dumped as j616c-fwdata-dump. Is that driver, or the dump procedure, available somewhere? I'd like to run it on this 40-core unit, with aura's GPU tables in the device tree, and send you the dump, or a 40-core profile built from it.

    🤖 Generated with Claude Code

  2. added a commit that references this issue on Nov 9, 2026
  3. DjDeveloperr commented on Nov 9, 2026

    @DjDeveloperr
    Contributor

    Thanks for the detailed report. The two remaining gaps have fixes merged now. They are built but not yet tested on a J614c, since there isn't one here, so please try the next build.

    1. Internal display (aurora-silicon/aura#132, #1345)

    You read it right: on the M4 Max, the kernel and aura disagreed, and the J616c's internal display was on simpledrm in the integration tree too. Both sides now work the way the J616s does:

    • Kernel tree: t6041-j614c-j616c.dtsi no longer enables /display-subsystem. It and the panel stay disabled for the boot loader's handoff.
    • Kernel gate: the M4 Max display gate first runs the same internal-display check as the M4 Pro. When the handoff is there and the DCP is running, it enables the DARTs, the mailbox, the display subsystem, the DCP and the panel. It then enables the HDMI and USB-C chains, which need that display subsystem.
    • aura: the M4 Max handoff entry serves apple,j614c as well as apple,j616c, as in your one-line change. Your log showed every ADT and hardware check passing on the J614c, so that entry's values are used as they are.

    With both in place, the aura log should end in DCP: T6041: handoff published (...) and dcpext0/dcpext1 firmware memory published, and the kernel should log apple-t6041-display: display handoff accepted. If it doesn't, the DCP: T6041: lines from the aura log and dmesg | grep -i -E 'apple-t6041-display|dcp' will show which check refused.

    One change in behaviour: without a display handoff, the external displays now stay off together with the internal one, as on the J616s.

    2. GPU on the 40-core chip with 25G83 (#1345)

    T6041 now has a 25G83 build that runs on the 40-core configuration. It reuses what the 25F84 40-core profile knows (four clusters of ten, the operand and deflake layouts, the compute pool, the hardware data and power words) with the 25G83 rules and the 25G83 register-map layout. That layout reproduces both 25G83 maps we know, the M4 Pro's and #1050's 32-core one, here with the 40-core chip's eight memory controllers. The core-count check is unchanged.

    One value can't be derived: the identity of the GPU firmware image your Mac loads. iBoot fills part of the loaded image for each Mac, so it differs between machines. The only T6041 25G83 image measured so far comes from #1050's 32-core J616C, and that's the one the build admits. Your J614c most likely loads a different one. If the GPU is still refused on the next build, please post the output of:

    sudo dmesg | grep -A3 'unsupported GPU firmware build'
    

    The sha256 in that line, together with its TEXT/DATA/entropy values, is all that's missing, and adding it is a one-line change. Your first boot's log already has that line if you still have it (journalctl -k -b <that boot> | grep -A3 'unsupported GPU firmware build').

    If the image is admitted and something fails after that, sudo dmesg | grep -i -E 'asahi|gpu|agx' from that boot is what we'd need.

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

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions