hang during SI VERDE init after UVD, truncated drm:amdgpu, on Mac

mwigzell · Yesterday 02:18:56

Machine: MacBookPro12,1, AMD 1002:6821 (VERDE / M370X class), Intel [8086:0d26] (rev 08)
Distro / kernel: 7.0.3-arch1-1. x86_64
apple_set_os.efi:Used own shim app to run it then run my systemd boot
Boot chain that matters: systemd-boot, no nomodeset, no amdgpu blacklist for the failing entry.
Initramfs note: kms removed from HOOKS — that caused amdgpu to run later, after root switch, thus serial console started working
boot options:loglevel=7 drm.debug=0x1e console=tty0 console=ttyUSB0,115200n8 earlyprintk=serial,ttyUSB0,115200 memmap=1M$0x47f500000 ramoops.mem_address=0x47f500000 ramoops.mem_size=0x100000
rescue boot options: nomodeset
ramoops: No entry
serial (tail):

[   27.671610] amdgpu: Virtual CRAT table created for CPU
[   27.672587] amdgpu: Topology: Add CPU node
[   27.673457] amdgpu 0000:01:00.0: SI support provided by amdgpu.
[   27.673457] Use radeon.si_support=1 amdgpu.si_support=0 to override.
[   27.675969] amdgpu 0000:01:00.0: initializing kernel modesetting (VERDE 0x1002:0x6821 0x106B:0x0149 0x83).
[   27.677285] amdgpu 0000:01:00.0: register mmio base: 0xB0C00000
[   27.678120] amdgpu 0000:01:00.0: register mmio size: 262144
[   27.678954] amdgpu 0000:01:00.0: detected ip block number 0 <common_v1_0_0> (si_common)
[   27.680061] amdgpu 0000:01:00.0: detected ip block number 1 <gmc_v6_0_0> (gmc_v6_0)
[   27.681143] amdgpu 0000:01:00.0: detected ip block number 2 <ih_v1_0_0> (si_ih)
[   27.682279] amdgpu 0000:01:00.0: detected ip block number 3 <gfx_v6_0_0> (gfx_v6_0)
[   27.683320] amdgpu 0000:01:00.0: detected ip block number 4 <sdma_v1_0_0> (si_dma)
[   27.684134] amdgpu 0000:01:00.0: detected ip block number 5 <smu_v6_0_0> (si_dpm)
[   27.684881] amdgpu 0000:01:00.0: detected ip block number 6 <dce_v1_0_0> (dm)
[   27.685811] amdgpu 0000:01:00.0: detected ip block number 7 <uvd_v3_1_0> (uvd_v3_1)
[   27.686579] amdgpu 0000:01:00.0: detected ip block number 8 <vce_v1_0_0> (vce_v1_0)
[   27.687351] amdgpu 0000:01:00.0: Fetched VBIOS from VFCT
[   27.688465] amdgpu 0000:01:00.0: [drm] ATOM BIOS: 113-C567A1-006
[   27.710239] amdgpu 0000:01:00.0: Found VCE firmware Version: 50.0 Binary ID: 1
[   27.711430] kfd kfd: VERDE  not supported in kfd
[   27.721236] Console: switching to colour dummy device 80x25
[   27.721725] amdgpu 0000:01:00.0: vgaarb: deactivate vga console
[   27.721732] amdgpu 0000:01:00.0: Trusted Memory Zone (TMZ) feature not supported
[   27.721861] amdgpu 0000:01:00.0: [drm] PCIE gen 3 link speeds already enabled
[   27.721868] amdgpu 0000:01:00.0: vm size is 64 GB, 2 levels, block size is 10-bit, fragment size is 9-bit
[   27.731687] amdgpu 0000:01:00.0: VRAM: 2048M 0x0000000000000000 - 0x000000007FFFFFFF (2048M used)
[   27.731699] amdgpu 0000:01:00.0: GART: 1024M 0x0000000000000000 - 0x000000003FFFFFFF
[   27.731717] amdgpu 0000:01:00.0: [drm] Detected VRAM RAM=2048M, BAR=256M
[   27.731722] amdgpu 0000:01:00.0: [drm] RAM width 128bits GDDR5
[   27.731891] amdgpu 0000:01:00.0:  2048M of VRAM memory ready
[   27.731898] amdgpu 0000:01:00.0:  7930M of GTT memory ready.
[   27.731912] amdgpu 0000:01:00.0: [drm] GART: num cpu pages 262144, num gpu pages 262144
[   27.732553] amdgpu 0000:01:00.0: [drm] PCIE GART of 1024M enabled (table at 0x0000000001400000).
[   27.749095] amdgpu 0000:01:00.0: [drm] Internal thermal controller without fan control
[   27.749115] amdgpu 0000:01:00.0: [drm] si dpm initialized
[   27.755867] amdgpu 0000:01:00.0: [drm] Found UVD firmware Version: 64.0 Family ID: 13
[  OK  ] Started D-Bus System Message Bus.
[  OK  ] Reached target Basic System.
         Starting Network Manager...
         Starting User Login Management...
[  OK  ] Started User Login Management.
[   28.066233] amdgpu 0000:01:00.0: [drm:amdgpu_r??

Question: Given this last "drm:amdgpu_r" (terminated by two reverse question marks) on this SI path, what is the next init step I should inspect, and is there a known SI/MacBook issue or a recommended single amdgpu/drm knob to try?
Further Steps:
I am not sure how I might attach the entire serial log if that is of use please tell me how.
I am happy to run one suggested experiment per boot and return serial.

Last edited by mwigzell (Yesterday 02:22:26)

seth · Yesterday 08:39:12

Does the radeon module sucessfully initialize the GPU?

"amdgpu.aspm=0 amdgpu.bapm=0 amdgpu.runpm=0 amdgpu.audio=0" I'd also try "amdgpu.dpm=0" but chances are that the boot will even fail more.
Randomly also try https://wiki.archlinux.org/title/AMDGPU … timed_out)

mwigzell · Today 03:06:46

radeon path:

Options: modprobe.blacklist=amdgpu video=simplefb:off mem_sleep_default=s2idle
Outcome:: Shell OK; amdgpu absent; radeon does not bind 01:00.0

Experiment 1

Optionsloglevel=7 drm.debug=0x1e console=tty0 console=ttyUSB0,115200n8 earlyprintk=serial,ttyUSB0,115200 memmap=1M$0x47f500000 ramoops.mem_address=0x47f500000 ramoops.mem_size=0x100000 amdgpu.aspm=0 amdgpu.bapm=0 amdgpu.runpm=0 amdgpu.audio=0
Outcome:Same result as above: truncated output : "[ 119.741153] amdgpu 0000:01:00.0: [drm:amdgpu_ri"

Experiment 2

Optionsloglevel=7 drm.debug=0x1e console=tty0 console=ttyUSB0,115200n8 earlyprintk=serial,ttyUSB0,115200 memmap=1M$0x47f500000 ramoops.mem_address=0x47f500000 ramoops.mem_size=0x100000 amdgpu.aspm=0 amdgpu.bapm=0 amdgpu.runpm=0 amdgpu.audio=0 amdgpu.dpm=0
Outcome:Same result a above: truncated output: "[ 35.433353] amdgpu 0000:01:00.0: [drm:amdgpu_ri"

Arch Linux

#1 Yesterday 02:18:56