CVE-2025-38205

What it actually does

Uninitialized dummy pitch value in populate_dummy_dml_surface_cfg() allows downstream divide-by-zero in CalculateVMAndRowBytes() within AMD display code path.

Why it matters to the business

Local denial of service via kernel panic or crash on systems using AMD GPUs, disrupting workloads until reboot; higher urgency for container-optimized environments given related privilege-escalation patterns in the ecosystem.

How an attack unfolds

An attacker with local access triggers the AMD display dummy surface configuration path without updated pitch values, causing integer division by zero that terminates the kernel process; observed in container-optimized OS privilege-escalation campaigns targeting similar Linux kernel flaws.

How to fix it

Apply the upstream fix commit 8044f981b2cf8c32fe1bd5d1fc991552cdf7ffe0 or later; update smarter-device-manager-fips to the fixed release; rebuild and deploy kernels containing the corrected populate_dummy_dml_surface_cfg() initialization.

How to confirm the fix worked

Confirm kernel source contains the commit 8044f981b2cf8c32fe1bd5d1fc991552cdf7ffe0; verify dummy pitch is initialized to 1 in populate_dummy_dml_surface_cfg(); test display mode-setting under load without triggering division faults.

Typical attack path

1. Obtain local user or container access on affected Linux kernel
2. Trigger AMD display initialization or mode-setting code paths that invoke populate_dummy_dml_surface_cfg()
3. Ensure dummy values remain uninitialized
4. Reach CalculateVMAndRowBytes() causing divide-by-zero fault
5. Kernel panic or crash occurs

How to detect it

Review kernel logs for divide-by-zero faults in drm/amd/display or CalculateVMAndRowBytes(); correlate with AMD GPU initialization events; watch for unexpected container escapes or privilege-escalation attempts listed under related Linux kernel CVEs.

• divide by zero in CalculateVMAndRowBytes
• populate_dummy_dml_surface_cfg
• drm/amd/display

What your security team should do

• Prioritize patching all AMD GPU systems within 30 days given 480-day exposure window
• Monitor for related container-optimized OS privilege-escalation activity
• Restrict local unprivileged access where possible until patched
• Include the fixed commit in all downstream kernel builds