AMD RX 7700 XT (Navi 32) GPU Freezes on Ubuntu 25.10 / Kernel 6.17 - Debugging Notes

README.md

AMD RX 7700 XT (Navi 32) GPU Freezes on Ubuntu 25.10 / Kernel 6.17

System Configuration

Component	Details
GPU	AMD Radeon RX 7700 XT / 7800 XT (Navi 32, gfx1101)
OS	Ubuntu 25.10
Desktop	KDE Plasma (Wayland)
Kernel	6.17.0-6-generic (issue present), 6.14.0-15-generic (testing)
CPU	Intel Core i7-12700F
RAM	64 GB
VRAM	12 GB GDDR6

Symptoms

• Full system lockups (hard freeze, requires power cycle)
• Random occurrence - not tied to specific GPU load
• Started after upgrading from Ubuntu 25.04 to 25.10 (kernel 6.14 → 6.17)
• No recovery possible once frozen (SysRq keys unresponsive)

Logs & Errors

Recurring DRM Scheduler Errors

[drm:drm_sched_entity_push_job [gpu_sched]] *ERROR* Trying to push to a killed entity

These appear frequently in syslog, often in pairs, throughout normal operation.

Display Manager IRQ Warning (Boot)

workqueue: dm_irq_work_func [amdgpu] hogged CPU for >10000us 4 times, consider switching to WQ_UNBOUND

SMU Driver Version Mismatch

amdgpu 0000:03:00.0: amdgpu: smu driver if version = 0x0000003d, smu fw if version = 0x00000040
amdgpu 0000:03:00.0: amdgpu: SMU driver if version not matched

Other Notable Messages

amdgpu 0000:03:00.0: amdgpu: DF poison setting is inconsistent(1:0:0:0)!
amdgpu 0000:03:00.0: amdgpu: Poison setting is inconsistent in DF/UMC(0:1)!
amdgpu: Freeing queue vital buffer 0x..., queue evicted

Kernel Parameters Tested

Current GRUB Configuration (kernel 6.17)

amdgpu.gpu_recovery=1
amdgpu.ppfeaturemask=0xfffd7fff
amdgpu.dpm=1
amdgpu.runpm=0
amdgpu.dc=1
amdgpu.dcfeaturemask=0x8
amdgpu.tmz=0
amdgpu.sg_display=0
amdgpu.dcdebugmask=0x10
iommu=soft

Previously Tested (kernel 6.14)

amdgpu.dpm=1
amdgpu.gpu_recovery=1
amdgpu.vm_fragment_size=9
amdgpu.lockup_timeout=60000
amdgpu.noretry=0
amdgpu.ppfeaturemask=0xffffffff

Debugging Steps Taken

1. Removed Custom Power Limit Services

Had two conflicting systemd services limiting GPU power:

• amdgpu-power-limit.service → 220W
• gpu-power-limit.service → 180W

Resolution: Removed both services, reset to default 200W power cap.

sudo systemctl disable --now amdgpu-power-limit.service gpu-power-limit.service
sudo rm /etc/systemd/system/amdgpu-power-limit.service /etc/systemd/system/gpu-power-limit.service
echo 200000000 | sudo tee /sys/class/drm/card1/device/hwmon/hwmon*/power1_cap

2. Downgraded to Kernel 6.14

Set kernel 6.14.0-15-generic as default boot option to test if issue is a 6.17 regression:

sudo sed -i 's|GRUB_DEFAULT=.*|GRUB_DEFAULT="gnulinux-advanced-UUID>gnulinux-6.14.0-15-generic-advanced-UUID"|' /etc/default/grub
sudo update-grub

Status: Testing in progress

Related Resources

• RDNA3 TLB Fence Issues Guide
• Arch Forums - System freeze because of amdgpu driver
• Arch Forums - AMD GPU crashing (SOLVED)
• freedesktop.org Bug #111481 - AMD Navi GPU frequent freezes

Known Issues (Kernel 6.14-6.17 + RDNA3)

Per community reports, there's a documented kernel bug in memory management affecting RX 7000 series GPUs on kernels 6.14-6.17, causing:

• GPU freezes
• System hangs
• "Fence timeout" errors in logs

Next Steps

1. Test stability on kernel 6.14
2. If stable on 6.14, monitor upstream for 6.17+ fixes
3. Consider testing amdgpu.gfx_off=0 parameter if issues persist
4. File bug report with AMD/kernel team if reproducible case identified

Hardware Details

amdgpu 0000:03:00.0: amdgpu: initializing kernel modesetting (IP DISCOVERY 0x1002:0x747E 0x1DA2:0x475F 0xFF)
VRAM: 12272M (12272M used)
GART: 512M
RAM width 192bits GDDR6
SE 3, SH per SE 2, CU per SH 10, active_cu_number 54

---

This gist documents ongoing debugging efforts for AMD RDNA3 GPU stability issues on recent Linux kernels. Information may help upstream developers working on compatibility fixes.

Generated with Claude Code - please verify all information independently.