iMac20,1 (Navi 14): amdgpu fails at SMU init, `hw_init of IP block <smu> failed -62`

[FjjDessoyCaraballo]

Summary

On a fleet of iMac20,1 machines running Ubuntu 22.04 with t2 kernels, amdgpu fails to initialise. The SMU never responds to the driver's first message, init aborts, no DRM device is created, and the machine boots headless. The GPU initialises correctly under macOS on the same hardware.

Environment

Model	iMac20,1
GPU	Navi 14 — Radeon RX 5500/5500M / Pro 5500M [1002:7340] at 0000:03:00.0
VBIOS	113-D322A1XL-011
OS	Ubuntu 22.04 (Jammy)
Kernel	6.12.81-2-t2-jammy (also reproduced on 6.6.72-1-t2-jammy)
linux-firmware	20220329.git681281e4-0ubuntu3.40 (navi14_smc.bin dated 2025-09-18)

Symptom

Boot proceeds normally until amdgpu probes. The driver deactivates the VGA console, then fails ~3s later and tears down, leaving no console output device. The screen freezes on the last drawn frame and never updates again.

The system is fully booted behind that stale image — SSH, DHCP, and cron all work normally. systemctl list-jobs reports no jobs running. It is not a hang.

ls /sys/class/drm/ shows only version — no card0.

Relevant log

amdgpu 0000:03:00.0: amdgpu: Fetched VBIOS from VFCT
amdgpu: ATOM BIOS: 113-D322A1XL-011
amdgpu 0000:03:00.0: amdgpu: reserve 0x900000 from 0x80fd000000 for PSP TMR
amdgpu 0000:03:00.0: amdgpu: RAS: optional ras ta ucode is not available
amdgpu 0000:03:00.0: amdgpu: RAP: optional rap ta ucode is not available
amdgpu 0000:03:00.0: amdgpu: SECUREDISPLAY: securedisplay ta ucode is not available
amdgpu 0000:03:00.0: amdgpu: use vbios provided pptable
amdgpu 0000:03:00.0: amdgpu: smc_dpm_info table revision(format.content): 4.5
amdgpu 0000:03:00.0: amdgpu: SMU: I'm not done with your previous command: SMN_C2PMSG_66:0x00000006 SMN_C2PMSG_82:0x00000000
amdgpu 0000:03:00.0: amdgpu: Failed to enable requested dpm features!
amdgpu 0000:03:00.0: amdgpu: Failed to setup smc hw!
[drm:amdgpu_device_init [amdgpu]] *ERROR* hw_init of IP block <smu> failed -62
amdgpu 0000:03:00.0: amdgpu: amdgpu_device_ip_init failed
amdgpu 0000:03:00.0: amdgpu: Fatal error during GPU init
WARNING: CPU: 4 PID: 426 at drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c:631 amdgpu_irq_put+0xa4/0xc0 [amdgpu]
amdgpu 0000:03:00.0: probe with driver amdgpu failed with error -62

PSP completes successfully; SMU is the first block to fail. Message 0x6 (EnableAllSmuFeatures) gets response 0x00000000 — no reply at all.

Possibly related ACPI errors early in boot:

ACPI: [Firmware Bug]: BIOS _OSI(Linux) query ignored
ACPI Error: Aborting method \_SB._OSC due to previous error (AE_AML_BUFFER_LIMIT)
ACPI Error: Aborting method \_SB.PR00._OSC due to previous error (AE_ALREADY_EXISTS)

Ruled out

• Hardware. macOS Internet Recovery boots and renders correctly on the same machine.
• Kernel version. Reproduced on both 6.6.72 and 6.12.81 t2 kernels.
• Firmware. navi14_smc.bin present and current (Sep 2025 blob).
• Retained SMU state. Full power cycle with PSU disconnected 30s — no change.
• Filesystem. fsck clean.
• Kernel parameters tried, no effect: amdgpu.dpm=0, amdgpu.ppfeaturemask=0, amdgpu.sg_display=0, and removing intel_iommu=on iommu=pt.
• PCI remove + rescan. Device re-enumerates cleanly but amdgpu does not re-probe.

Current kernel cmdline

BOOT_IMAGE=/boot/vmlinuz-6.12.81-2-t2-jammy root=UUID=... ro
pcie_ports=compat acpi=force intel_iommu=on iommu=pt
kvm.enable_virt_at_load=0 systemd.show_status=true

Fleet context

Multiple identically-imaged iMac20,1 units. Behaviour progressed from working, to booting successfully roughly half the time, to never booting to a display. A 2021 Arch Linux forum report on the same model and same PCI address describes the same ~50% success rate and the same -62, with amdgpu.dpm=0 as a workaround on kernel 5.11 that stopped working on later kernels.

Questions

1. Is amdgpu known-working on iMac20,1 under any t2 kernel? If so, which, and with what parameters?
2. Given the GPU initialises under macOS, is this an OS-identity issue — i.e. does apple_set_os / rEFInd macOS spoofing / acpi_osi=Darwin apply to iMacs, or is that MacBook-only?
3. Are the _OSC ACPI aborts significant here, and is acpi=force advisable on this model?
4. Is there a known-good amdgpu.ppfeaturemask value for this Apple VBIOS, along the lines of the iMac17,1 workaround?

Happy to run diagnostics, test patches, or supply full dmesg / acpidump on request.

[iurjscsi1101500]

1. Yes it working on 5300, using this patch: https://github.com/t2linux/linux-t2-patches/blob/main/6001-drm-amd-pm-Fix-boot-problems-in-5300.patch
2. No, its only related to amdgpu
3. Same as 2, the problem is only related to amdgpu
4. No, the ppfeaturemask stuff runs after the code, where the problem occurs

can you give ioreg -l from macOS for that model, then ill release a patch and you can use that then

[FjjDessoyCaraballo]

Here it is

ioreg.txt

[AdityaGarg8]

7.1.x kernels have the patch @iurjscsi1101500 told about. You seem to be on 6.18

[FjjDessoyCaraballo]

Built 7.1.5 with 6001 applied (verified 0xABC9AFBB at smu_v11_0.c:681). Still fails, note the failure is at msg_reg: 6 — after the two SetAllowedFeaturesMask calls your patch replaces. Suggests the masks are being accepted and the failure is at EnableAllSmuFeatures, which would fit the hardcoded values being 5300-specific. The ioreg I sent should have the right ones for this variant.

[iurjscsi1101500]

use 0xABC9ACA3 at 681 (low) and 0xFFFFFD42 for high