CVE-2026-46263: Linux AMD Display Driver Out-of-Bounds Array Access

Technical summary

CVE-2026-46263 is an out-of-bounds array access vulnerability in drivers/gpu/drm/amd/display/dc/resource/dcn*/dcn*_resource.c. The stream_encoder_create() function uses an engine ID (eng_id) to directly index the stream_enc_regs[] array, which contains only 5 elements (indices 0–4). The original code checks eng_id <= ENGINE_ID_DIGF (where ENGINE_ID_DIGF = 5), permitting an out-of-bounds access at index 5. The fix adds an explicit bounds check using ARRAY_SIZE() before indexing, rejecting invalid engine IDs. This is a classic CWE-125 (out-of-bounds read) that can also lead to memory corruption depending on the kernel layout and call context.

Business impact

Organizations running vulnerable Linux kernels with AMD GPU drivers face elevated risk of local privilege escalation and information disclosure. A local attacker with user-level access could read kernel memory, leak cryptographic keys, or corrupt kernel state to gain root privileges. This is particularly critical for multi-tenant environments, containerized workloads, and shared GPU resources where untrusted code may execute alongside trusted applications.

Affected systems

The vulnerability affects Linux kernels with AMD display driver support, specifically the DCN (Display Core Next) code paths used in modern AMD GPUs. Systems running kernels before the fix was integrated are vulnerable. Desktop, laptop, and workstation systems with AMD Radeon or Radeon Pro GPUs running vulnerable kernel versions are at risk. Affected kernel versions should be verified against your distribution's security advisories and the kernel.org CVE tracker.

Exploitability

Exploitability is moderate-to-high. The flaw requires local access and user-level privileges (no special capabilities needed initially), with a straightforward trigger—calling the stream encoder creation function with an out-of-range engine ID. However, practical exploitation depends on the attacker's ability to control the parameters passed to the vulnerable code path, which may be limited by GPU driver initialization logic. The attack surface is active during GPU setup and driver load, making it viable on systems where untrusted code can interact with GPU drivers or where GPU passthrough is used in VMs.

Remediation

Update the Linux kernel to a version that includes the bounds-check fix. Most major distributions (Ubuntu, RHEL, Debian, SUSE, Fedora) will backport the fix to their supported kernel branches. Verify the specific patch version for your distribution by consulting their security advisories. For systems unable to patch immediately, disable GPU driver loading for untrusted user accounts or restrict GPU access via access controls if possible. Alternatively, disabling CONFIG_DRM_AMD_DC at compile time removes the vulnerable code entirely but disables AMD display support.

Patch guidance

Patch versions vary by distribution; verify the exact kernel version from your vendor's security advisory. The upstream kernel fix involves adding an explicit ARRAY_SIZE() bounds check before the stream_enc_regs[] indexing statement in dcn*_stream_encoder_create(). Apply security updates as released by your Linux distribution's maintainers. Kernel reboot is required after patching.

Detection guidance

Monitor system logs for GPU driver initialization errors or crashes related to stream encoder creation. Audit access to /dev/dri/* and GPU ioctl calls on systems with AMD GPUs. Use kernel Address Sanitizer (KASAN) or similar debugging tools in test environments to catch out-of-bounds accesses. In production, look for unexpected kernel warnings or panics mentioning dcn*_resource.c or stream_encoder. Container orchestration platforms should ensure host kernels are patched before GPU device passthrough is permitted to containers.

Why prioritize this

This vulnerability merits prompt patching due to its HIGH CVSS 3.1 score (7.8), local attack vector requiring only user privileges, and the high impact on confidentiality, integrity, and availability. It enables privilege escalation and information disclosure without user interaction. Systems with GPU workloads or multi-tenant configurations should prioritize this fix within their patch cycle.

Risk score, explained

CVSS 3.1 score of 7.8 (HIGH) reflects: Attack Vector Local (AV:L), low attack complexity (AC:L), low privilege requirement (PR:L), no user interaction (UI:N), unchanged scope (S:U), and high impact across confidentiality, integrity, and availability (C:H/I:H/A:H). The local-only attack vector prevents remote exploitation, but the low privilege bar and direct path to kernel memory access elevate the score significantly.

Frequently asked questions