[BUG] amdgpu NULL page fault causing gfx ring timeout crash on Radeon 8060S (RDNA4) — reproducible across multiple kernel versions

["Seamus Na" <slamdunk111@163.com>]

[-- Attachment #1: Type: text/plain, Size: 3317 bytes --]

Attention: This bug bothers me almost 1 week!  I bought this AMD 395 hardware for better AI inference, but untill now I Still cannot use AI Inference because it crashes all the time;
Could the AMD Officers you give me back $1500 for AI not working at all??


AMD Linux Kernel Driver Bug Report
===================================


System: AMD Ryzen AI MAX+ 395 w/ Radeon 8060S (Strix Halo, RDNA4 iGPU)
GPU: 0000:66:00.0, PCI ID 1002:1586, rev c1
OS: Ubuntu 24.04.4 LTS
Kernel: 7.0.0-28-generic (also reproduced on 6.14.0-1020-oem and 6.17.0-1030-oem)
amdgpu driver: kernel built-in (no external DKMS)
Xorg: xserver-xorg-video-amdgpu 23.0.0
Boot params: amdgpu.gttsize=81920 amdgpu.vm_size=128 amdgpu.vm_update_mode=3


PROBLEM
-------
System crashes due to amdgpu driver NULL page fault. The GPU attempts to access
memory address 0x0, triggering a gfxhub page fault with PERMISSION_FAULTS=0x3.
This causes gfx ring timeout, device wedged state, and system crash requiring
hard reboot.


KEY ERROR LOG
-------------
amdgpu 0000:66:00.0: [gfxhub] page fault (src_id:0 ring:24 vmid:7 pasid:17)
amdgpu 0000:66:00.0: Process Xorg pid 2696 thread Xorg:cs0 pid 3080
amdgpu 0000:66:00.0: in page starting at address 0x0000000000000000 from client 10
amdgpu 0000:66:00.0: GCVM_L2_PROTECTION_FAULT_STATUS:0x00701431
amdgpu 0000:66:00.0: Faulty UTCL2 client ID: SQC (data) (0xa)
amdgpu 0000:66:00.0: PERMISSION_FAULTS: 0x3
amdgpu 0000:66:00.0: MAPPING_ERROR: 0x0
amdgpu 0000:66:00.0: RW: 0x0


amdgpu 0000:66:00.0: ring gfx_0.0.0 timeout, signaled seq=1, emitted seq=3
amdgpu 0000:66:00.0: Starting gfx_0.0.0 ring reset
amdgpu 0000:66:00.0: Ring gfx_0.0.0 reset succeeded
amdgpu 0000:66:00.0: [drm] device wedged, but recovered through reset
amdgpu 0000:66:00.0: ring gfx_0.0.0 timeout, signaled seq=172, emitted seq=174
amdgpu 0000:66:00.0: Starting gfx_0.0.0 ring reset
amdgpu 0000:66:00.0: Ring gfx_0.0.0 reset succeeded


REPRODUCTION
------------
1. Boot with any kernel supporting RDNA4 (tested: 6.14 OEM, 6.17 OEM, 7.0 generic)
2. Start Xorg desktop session (GNOME)
3. Use any GPU-rendering application (even just desktop compositing)
4. NULL page fault occurs in Xorg GPU context (PASID 17)
5. gfx ring timeout → crash


CRASH FREQUENCY
---------------
20+ kernel crashes since Aug 3. Increasing frequency from 1-2/day to 6+/day.
Also affects rocm-smi and python/ROCm inference processes independently.


ADDITIONAL ISSUES
-----------------
Chronic amdgpu workqueue hogging observed since July 28:
- svm_range_restore_work: peak 259 hog events (July 31)
- amdgpu_amdkfd_restore_userptr_worker: peak 67 hog events (Aug 1)
- Fence fallback timer expired on ring comp_1.2.1 (July 29)


TIMELINE
--------
- Jul 24 - Aug 2: Normal operation, zero crashes with ROCm inference
- Aug 3: First crash (same day as physical relocation BJ→SH)
- Aug 6-8: Frequent crashes (3-6+ per day)


FULL DETAILED REPORT
--------------------
https://github.com/slamdunk111/amd-gpu-bugs/issues/1


Crash dumps available at /var/crash/ on the affected system.


Please investigate whether this is a known issue with RDNA4 iGPU support in the
current amdgpu kernel driver. A fix or workaround is urgently needed as the
system is currently unusable for any GPU workload.


--
nawenyu (gmk)
AMD Ryzen AI MAX+ 395 w/ Radeon 8060S, Ubuntu 24.04.4


[-- Attachment #2: Type: text/html, Size: 5836 bytes --]