CVE-2026-64516 Overview
CVE-2026-64516 is a Linux kernel vulnerability in the AMD GPU (amdgpu) driver, specifically within the Video Compression Engine 1 (VCE1) subsystem. The flaw resides in the handling of the VCPU Buffer Object (BO), where the firmware offset was not incorporated into the VCPU BO size calculation. This oversight enables out-of-bounds access during firmware loading operations. A local authenticated attacker with low privileges can leverage the flaw to compromise confidentiality, integrity, and availability of the affected system. The issue has been resolved upstream through commit c16fe59f622a080fc457a57b3e8f14c780699449.
Critical Impact
Local low-privileged attackers can trigger out-of-bounds memory access in the AMD GPU driver, potentially leading to kernel memory corruption and full system compromise.
Affected Products
- Linux kernel versions containing the vulnerable drm/amdgpu/vce1 driver code
- Systems using AMD GPUs with VCE 1 hardware
- Distributions shipping affected kernel builds prior to the backport commits
Discovery Timeline
- 2026-07-25 - CVE-2026-64516 published to NVD
- 2026-07-27 - Last updated in NVD database
Technical Details for CVE-2026-64516
Vulnerability Analysis
The vulnerability affects the AMD GPU kernel driver's VCE 1 firmware initialization path. The VCPU BO is a memory region allocated to hold the firmware microcode used by the VCE hardware block. The firmware content resides at a specific offset within the BO, but the driver did not subtract this offset from the total firmware size when validating buffer boundaries. This mismatch produces an incorrect effective size, allowing memory accesses that extend beyond the allocated region.
The upstream fix introduces three corrections. First, the firmware size calculation now subtracts the offset within the VCPU BO to prevent out-of-bounds reads or writes. Second, stack and data offsets are aligned to the 32K Translation Lookaside Buffer (TLB) size to satisfy hardware alignment requirements. Third, the driver now validates that the firmware microcode fits within the reserved space before proceeding.
Root Cause
The root cause is an out-of-bounds access (CWE-787/CWE-125) resulting from incorrect size arithmetic in the VCE 1 firmware loading routine. The VCPU BO allocation did not account for the internal firmware offset, and no upfront validation confirmed that the microcode fit within the reserved buffer. Missing TLB alignment for stack and data offsets compounded the boundary handling defect.
Attack Vector
Exploitation requires local access with low privileges on a system with an AMD GPU exposing the VCE 1 interface. An attacker can interact with the affected driver code path through standard graphics device interfaces. The vulnerability yields a scope change, meaning the impact extends beyond the driver context into kernel memory. Successful triggering can produce kernel memory corruption, denial of service, or privilege escalation. Detailed exploitation technique is not publicly documented at time of writing; refer to the upstream kernel commit for the code-level fix.
Detection Methods for CVE-2026-64516
Indicators of Compromise
- Unexpected kernel oops or panic messages referencing amdgpu or vce_v1_0 functions in dmesg output
- Kernel Address Sanitizer (KASAN) reports indicating out-of-bounds access within the amdgpu driver
- Anomalous crashes or GPU resets on systems with AMD VCE 1 hardware following untrusted local process activity
Detection Strategies
- Inventory Linux endpoints and identify kernel versions predating the patch commits 3e5a1d5bb2ff061e64c7992f8e5404dfd4c2d0f3 and ce0de178ef08408f6ba8f2e9a13bf52fbe5852f4
- Monitor kernel logs for repeated crashes originating in the amdgpu VCE code path
- Correlate GPU driver faults with local process execution to identify potential exploitation attempts
Monitoring Recommendations
- Ingest dmesg and journald kernel logs into a centralized logging platform for pattern detection
- Enable KASAN on test and canary systems to surface memory safety violations before production impact
- Track running kernel versions across the fleet and alert when unpatched builds appear
How to Mitigate CVE-2026-64516
Immediate Actions Required
- Apply the vendor-supplied kernel update containing commits 3e5a1d5bb2ff061e64c7992f8e5404dfd4c2d0f3 and ce0de178ef08408f6ba8f2e9a13bf52fbe5852f4 as soon as your distribution publishes them
- Reboot affected systems after the update to load the patched kernel
- Restrict local access on multi-user systems and shared workstations that use AMD GPUs with VCE 1
Patch Information
The upstream fix was cherry-picked from commit c16fe59f622a080fc457a57b3e8f14c780699449. Stable-tree backports are available in the first kernel commit and the second kernel commit. Consult your Linux distribution's security tracker for packaged kernel updates.
Workarounds
- Blacklist the amdgpu module on systems where AMD GPU functionality is not required, using /etc/modprobe.d/ configuration
- Restrict access to graphics device nodes such as /dev/dri/* to trusted users only
- Limit local shell access on affected hosts until patched kernels are deployed
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
