You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
aurora-sep: M3 Max (J514C/J516C) boot.bin is written with chosen.asahi,t6031-gpu=0, so the G15 driver does not start on a normal boot #189
install-aurora-sep.sh gives the M3 Max one boot loader variant, t6031-gpu-off-1, written
with M3_MAX_SAFE_SWITCHES = chosen.asahi,t6031-dcp=1 chosen.asahi,t6031-gpu=0
(install-aurora-sep.sh:639-641; the reason is in the comment at :624-637: the GPU identity
read powers the GPU on every boot, and a stalled read would hang every boot of that
boot.bin). Only the M3 Max test kit switches t6031-gpu=1 on, for its own stage.
With #177 and #182 the kernel side is ready on the J516C. The G15 driver boots and passes the
acceptance checks, full dEQP and the concurrency tests when the boot loader fills the GPU node.
But a normal boot from the installed boot.bin leaves /soc/gpu@40a400000 disabled, and since #182 the driver says so once:
asahi: G15: native G15 not started: the boot loader left the GPU node /soc/gpu@40a400000 disabled (/chosen/asahi,t6031-gpu = "0")
Needed: a decision on when the installer writes t6031-gpu=1 for the M3 Max (for example,
after the kernel and Mesa that use it are installed, with the variant renamed), and how a
machine can go back if a boot hangs. Measured on a J516C: test boots with t6031-gpu=1
(chainloaded) on the #177/#182 kernels; no boot-time hang in the boots of 2026-10-22/23.
Nothing tested on the J514C.
With the candidate m1n1 in omacom/m1n1#19 (stage 2 5d17a034, package m1n1-aurora 1.6.1.aurora22-1, not merged
or released yet), m1n1 hands over the J516C display and GPU by default: chosen.asahi,t6031-dcp and chosen.asahi,t6031-gpu count as 1 unless a config line sets them. With the GPU switch on, m1n1 powers the GPU
for a short identity read on every boot. If that read ever stalled, every boot of that boot.bin would stop before
Linux. Naeem's J516C boots with t6031-gpu=1 have not hung, but this is the way back.
If Linux won't boot: from macOS Recovery
Shut the Mac down. Press and hold the power button until "Loading startup options" appears, then choose
Options → Continue.
Open Utilities → Terminal.
Run diskutil list. Find the 500 MB EFI partition of the Linux install (named like EFI - OMARC or EFI - ASAHI), for example disk0s4.
Mount it, go to its m1n1 folder and keep a copy of what is there (use Tab to complete the volume name):
diskutil mount disk0s4
cd "/Volumes/EFI - OMARC/m1n1"
cp boot.bin boot.bin.before-recovery
Add one line, either A or B below, then leave the folder, unmount the partition and restart:
cd /
diskutil unmount disk0s4
reboot
A. GPU off, display handoff kept
Use this when boot.bin is the one update-m1n1 wrote, which is the case unless someone edited it by hand.
The last command should print exactly chosen.asahi,t6031-gpu=0. At this candidate, that suppresses the GPU identity
read and leaves the GPU node disabled; the display handoff keeps its default.
Do not add a newline in front of the line. update-m1n1 ends every config line it writes with a newline, so none
is needed, and a blank line would make m1n1 treat the config as damaged (that gives the result of B).
This assumes stage 1 does not pass on its own chosen.asahi,t6031-gpu line. Stage 1's lines are read after
boot.bin's, and the last line for a name wins. A normal install's stage 1 passes on only its version and the
EFI partition. If the Mac still hangs at the same place, use B.
B. Every M3 graphics handoff off
Use this if A did not help, or if boot.bin may have been edited by hand.
The last command should print m1n1.config-lost=1. m1n1 then treats the config as having lost a line. On the J516C
that drops every asahi,t6031- setting, from the owner, the defaults or stage 1, so there is no display or GPU
handoff and the loader leaves the boot framebuffer in place. No later line can switch it back. The leading newline is
deliberate: it keeps the marker on a line of its own even if the file did not end with one.
If neither helps
Put back a boot.bin you know booted, for example a copy you made yourself when the Mac last worked: cp <that file> boot.bin.
Don't take boot.bin.old on trust. update-m1n1 keeps only the boot.bin it last replaced, so after a later kernel
update it can already hold the new m1n1. On a Mac installed with install-aurora-sep.sh, the script's boot.bin.before-<version> is the boot loader from before that run. boot.bin.before-recovery from step 4 is
the boot.bin you just changed, not a working one.
Once Linux boots: choose the switches and rebuild
update-m1n1 rebuilds boot.bin from /etc/m1n1.conf. Both recovery lines above go away at the next rebuild: A's
because it isn't in the file, B's because update-m1n1 passes on only chosen.*, display= and mitigations=
lines. So put what you want in the file first:
With linux-aurora#246 the kernel GPU waits for the native display, so t6031-dcp=0 on its own still leaves m1n1's
GPU identity read on. For a GPU problem, use t6031-gpu=0.
printf '\nchosen.asahi,t6031-gpu=0\n' | sudo tee -a /etc/m1n1.conf
sudo update-m1n1
Notes:
The leading newline is fine in this file. update-m1n1 skips blank lines, and the newline finishes a last line
that had none (update-m1n1 reads line by line and skips an unterminated last line).
The switches are independent, and an owner's =0 wins over the default.
chosen.asahi,m3-defaults=0 turns the generated defaults off but keeps any line you wrote.
On a Mac installed with install-aurora-sep.sh, keep your lines outside the script's # >>> aurora-sep: …
block; the script rewrites only its own block.
Macs installed with install-aurora-sep.sh
Plain run: on integration/aurora (937b727e), the script keeps an M3 Max kernel-only
(is_m3_kernel_only_chip, line 6348) and freezes update-m1n1:
m3_freeze, lines 871-884, adds a # >>> aurora-sep: keep this M3's boot.bin as it is block with M1N1_UPDATE_DISABLED=1 to /etc/default/update-m1n1.
This was checked on the source only: 5d17's own payload parser ran on the real 5d17 boot.bin bytes (with the
J516C device tree) with each of these lines appended, as a J516C, including a stage 1 that passes on t6031-gpu=1.
No M3 Max has booted this stage 2 yet.
Source pins: m1n1 5d17a034, installer script 937b727e. These are recovery instructions for the candidate, checked with its real host payload parser, not a newly performed Mac Recovery test. This answers the recovery part; the existing installer variant and its migration decision remain open.
Update: this looks fixed on current installs by omacom/m1n1#19 (merged 28 Oct, merge commit bc79acc4).
The fix: m1n1 now hands the J516C display and GPU over by default, so a normal boot no longer needs t6031-gpu=1 written by hand.
Seen on other J516Cs: MikeK's Mac runs m1n1 stage 2 bc79acc4 with kernel 7.2.11-11-1-ARCH. The G15 driver starts there and reaches the firmware check, which would not happen with the GPU node left disabled. digital-brew's Mac (stage 2 77f000bf) shows the same.
Not checked: I have not tested a plain boot on my own J516C with this m1n1, only through lab test boots.
Activity
chriskearney commentedon Oct 28, 2026
With the candidate m1n1 in omacom/m1n1#19 (stage 2
5d17a034, packagem1n1-aurora 1.6.1.aurora22-1, not mergedor released yet), m1n1 hands over the J516C display and GPU by default:
chosen.asahi,t6031-dcpandchosen.asahi,t6031-gpucount as1unless a config line sets them. With the GPU switch on, m1n1 powers the GPUfor a short identity read on every boot. If that read ever stalled, every boot of that boot.bin would stop before
Linux. Naeem's J516C boots with
t6031-gpu=1have not hung, but this is the way back.If Linux won't boot: from macOS Recovery
Shut the Mac down. Press and hold the power button until "Loading startup options" appears, then choose
Options → Continue.
Open Utilities → Terminal.
Run
diskutil list. Find the 500 MBEFIpartition of the Linux install (named likeEFI - OMARCorEFI - ASAHI), for exampledisk0s4.Mount it, go to its m1n1 folder and keep a copy of what is there (use Tab to complete the volume name):
Add one line, either A or B below, then leave the folder, unmount the partition and restart:
A. GPU off, display handoff kept
Use this when boot.bin is the one update-m1n1 wrote, which is the case unless someone edited it by hand.
The last command should print exactly
chosen.asahi,t6031-gpu=0. At this candidate, that suppresses the GPU identityread and leaves the GPU node disabled; the display handoff keeps its default.
is needed, and a blank line would make m1n1 treat the config as damaged (that gives the result of B).
chosen.asahi,t6031-gpuline. Stage 1's lines are read afterboot.bin's, and the last line for a name wins. A normal install's stage 1 passes on only its version and the
EFI partition. If the Mac still hangs at the same place, use B.
B. Every M3 graphics handoff off
Use this if A did not help, or if boot.bin may have been edited by hand.
The last command should print
m1n1.config-lost=1. m1n1 then treats the config as having lost a line. On the J516Cthat drops every
asahi,t6031-setting, from the owner, the defaults or stage 1, so there is no display or GPUhandoff and the loader leaves the boot framebuffer in place. No later line can switch it back. The leading newline is
deliberate: it keeps the marker on a line of its own even if the file did not end with one.
If neither helps
Put back a boot.bin you know booted, for example a copy you made yourself when the Mac last worked:
cp <that file> boot.bin.Don't take
boot.bin.oldon trust. update-m1n1 keeps only the boot.bin it last replaced, so after a later kernelupdate it can already hold the new m1n1. On a Mac installed with
install-aurora-sep.sh, the script'sboot.bin.before-<version>is the boot loader from before that run.boot.bin.before-recoveryfrom step 4 isthe boot.bin you just changed, not a working one.
Once Linux boots: choose the switches and rebuild
update-m1n1 rebuilds boot.bin from
/etc/m1n1.conf. Both recovery lines above go away at the next rebuild: A'sbecause it isn't in the file, B's because update-m1n1 passes on only
chosen.*,display=andmitigations=lines. So put what you want in the file first:
chosen.asahi,t6031-gpu=0.chosen.asahi,t6031-dcp=0.With linux-aurora#246 the kernel GPU waits for the native display, so
t6031-dcp=0on its own still leaves m1n1'sGPU identity read on. For a GPU problem, use
t6031-gpu=0.Notes:
that had none (update-m1n1 reads line by line and skips an unterminated last line).
=0wins over the default.chosen.asahi,m3-defaults=0turns the generated defaults off but keeps any line you wrote.install-aurora-sep.sh, keep your lines outside the script's# >>> aurora-sep: …block; the script rewrites only its own block.
Macs installed with install-aurora-sep.sh
937b727e), the script keeps an M3 Max kernel-only(
is_m3_kernel_only_chip, line 6348) and freezes update-m1n1:m3_freeze, lines 871-884, adds a# >>> aurora-sep: keep this M3's boot.bin as it isblock withM1N1_UPDATE_DISABLED=1to/etc/default/update-m1n1.in place, so the Mac gets none of EFI history cleanup: tolerate module-header symlinks while protecting boot references #19's defaults.
--m3-handoffvariant: this ist6031-gpu-off-1(lines 638-641). It writeschosen.asahi,t6031-dcp=1and
chosen.asahi,t6031-gpu=0, and EFI history cleanup: tolerate module-header symlinks while protecting boot references #19 honours that0. These Macs keep the GPU off until the script changes.This was checked on the source only: 5d17's own payload parser ran on the real
5d17boot.bin bytes (with theJ516C device tree) with each of these lines appended, as a J516C, including a stage 1 that passes on
t6031-gpu=1.No M3 Max has booted this stage 2 yet.
Source pins: m1n1
5d17a034, installer script937b727e. These are recovery instructions for the candidate, checked with its real host payload parser, not a newly performed Mac Recovery test. This answers the recovery part; the existing installer variant and its migration decision remain open.malik-na commentedon Oct 31, 2026
Update: this looks fixed on current installs by omacom/m1n1#19 (merged 28 Oct, merge commit
bc79acc4).t6031-gpu=1written by hand.bc79acc4with kernel7.2.11-11-1-ARCH. The G15 driver starts there and reaches the firmware check, which would not happen with the GPU node left disabled. digital-brew's Mac (stage 277f000bf) shows the same.125f0497is refused (G15: T6031 (M3 Max, J516C) GPU firmware TEXT not pinned: a third J516C build is refused #289). That is a separate problem from this one.I would close this unless someone still sees
t6031-gpu = "0"on a current install.