Use s2idle on T2 Macs with a discrete GPU

[T1mmsquared]

install/hardware/apple/fix-t2.sh detects T2 Macs by PCI ID (106b:180[12]) and gives every one of them the same kernel command line, including mem_sleep_default=deep. That works on integrated-only models, but a T2 Mac with a discrete GPU can never complete a deep suspend.

On a MacBookPro16,1 (Radeon Pro 5500M), every suspend attempt fails and falls back:

PM: suspend entry (deep)
amdgpu 0000:03:00.0: GPU mode1 reset failed
amdgpu 0000:03:00.0: PM: pci_pm_suspend_noirq(): amdgpu_pmops_suspend_noirq [amdgpu] returns -22
amdgpu 0000:03:00.0: PM: dpm_run_callback(): pci_pm_suspend_noirq returns -22
amdgpu 0000:03:00.0: PM: failed to suspend noirq: error -22
PM: noirq suspend of devices failed
PM: suspend exit
PM: suspend entry (s2idle)   <- systemd retries, this one succeeds

Counted over the journal on this machine: 3 deep attempts, 3 noirq suspend of devices failed, 3 GPU mode1 reset failed — a 100% failure rate. The machine still sleeps, because systemd retries with s2idle, so the symptom is easy to miss.

Two costs beyond the wasted transition:

1. Every lid close spends ~3 seconds suspending and then unwinding the entire device tree before the retry.
2. The failed pass tears down and recreates four Apple EFI variables each cycle — AppleSecureBootFailureReason, AppleSecureBootKernelFailureReason, AppleSecureBootWindowsFailureReason, AppleFailedBootVolumeUUID — so every suspend/resume writes NVRAM on hardware where that wear is not free.

Approach

mem_sleep_default=deep was added deliberately in #6562, so this does not remove it. It selects the sleep state from the GPU configuration using the existing omarchy-hw-hybrid-gpu helper:

if omarchy-hw-hybrid-gpu; then
  sleep_state="s2idle"
else
  sleep_state="deep"
fi

Integrated-only T2 Macs produce a byte-identical config file to today, so #6562's fix is preserved exactly where it works. Only discrete-GPU models change, and they change to the state the kernel was already falling back to on its own.

Included

• install/hardware/apple/fix-t2.sh — pick the sleep state from the GPU configuration
• migrations/1788072498.sh — move existing discrete-GPU installs off deep and rebuild the boot image, following the repair-marker pattern from migrations/1785944594.sh so a pre-reboot rerun stays idempotent
• test/shell.d/t2-hardware-test.sh — a stubbed omarchy-hw-hybrid-gpu, plus coverage for the discrete-GPU repair, its idempotency, the integrated-only path that must keep deep, and non-T2 hardware

Testing

• test/shell.d/t2-hardware-test.sh — 10/10 pass (6 pre-existing, 4 new)
• ./test/all — 4 failures, all pre-existing and environmental on this machine: three need a sibling omarchy-pkgs checkout that isn't present, one is a network-dependent theme URL check. None touch T2, suspend, or migrations.
• Verified on the affected hardware: MacBookPro16,1, Omarchy 4.0.1, linux-t2 7.1.8. After the change /sys/power/mem_sleep reads [s2idle] deep and the failing deep pass is gone.

Scope of the evidence

I can only confirm this directly on a MacBookPro16,1. The other T2 models with discrete GPUs — MacBookPro15,1 / 15,3 / 16,4, iMacPro1,1, MacPro7,1 — should be affected by the same amdgpu failure, but I have not tested them. Happy to adjust the detection if a T2 owner reports otherwise.