CVE-2026-16280: GPU MMU Information Disclosure Vulnerability

CVE-2026-16280 Overview

CVE-2026-16280 is an integer overflow vulnerability in the GPU driver's handling of sparse Physical Memory Regions (PMRs). When calculating physical offsets for sparse PMRs larger than 4 GB, address computations undergo 32-bit truncation. This flaw causes incorrect GPU Memory Management Unit (MMU) mappings. A non-privileged user can leverage the truncation to trigger access to unintended physical memory. Successful exploitation results in memory corruption or information disclosure. The vulnerability is tracked under CWE-190: Integer Overflow or Wraparound.

Critical Impact
A non-privileged local user can trigger unintended physical memory access through GPU MMU mappings, leading to memory corruption or disclosure of sensitive kernel data.

Affected Products

Imagination Technologies GPU driver (sparse PMR handling component)
Systems using affected GPU driver versions with sparse PMR allocations exceeding 4 GB
Refer to the Imagination Technologies Security Advisory for specific product versions

Discovery Timeline

2026-07-24 - CVE-2026-16280 published to NVD
2026-07-28 - Last updated in NVD database

Technical Details for CVE-2026-16280

Vulnerability Analysis

The vulnerability resides in the GPU driver code responsible for computing physical offsets within sparse PMRs. Sparse PMRs represent GPU-accessible memory regions that may be backed non-contiguously. When the driver calculates a physical offset for a PMR exceeding 4 GB, the arithmetic uses a 32-bit integer type. Values that require more than 32 bits are silently truncated. The truncated offset is then used to construct GPU MMU page table entries. As a result, the GPU maps virtual addresses to physical pages that do not correspond to the intended backing memory.

Root Cause

The root cause is the use of a 32-bit arithmetic type for offset computations that can legitimately exceed 0xFFFFFFFF when PMR sizes surpass 4 GB. The multiplication or addition of page indices with page sizes overflows the 32-bit boundary before the result is widened to a 64-bit physical address. This classifies as [CWE-190: Integer Overflow or Wraparound].

Attack Vector

A non-privileged user process interacts with the GPU driver through standard user-space interfaces to allocate a sparse PMR larger than 4 GB. The user then submits GPU work referencing offsets in the upper region of the PMR. The driver constructs MMU mappings using the truncated offsets. GPU commands issued by the user read from or write to physical pages outside the intended allocation. This yields memory corruption of kernel or other process memory, or discloses sensitive contents to the attacker.

No verified proof-of-concept code is publicly available. Refer to the Imagination Technologies Security Advisory for further technical detail.

Detection Methods for CVE-2026-16280

Indicators of Compromise

Unexpected GPU MMU fault entries in kernel logs referencing sparse PMR allocations
Non-privileged processes allocating unusually large (greater than 4 GB) sparse GPU memory regions
Kernel memory corruption symptoms coinciding with GPU workload execution

Detection Strategies

Audit GPU driver ioctl calls for sparse PMR allocations that exceed the 4 GB threshold
Correlate GPU page fault telemetry with process-level activity to identify anomalous access patterns
Monitor for user-space processes issuing large numbers of sparse mapping requests followed by GPU command submissions

Monitoring Recommendations

Enable verbose GPU driver logging on affected platforms and forward logs to a centralized SIEM for analysis
Track kernel oops and panic events tied to GPU driver stack traces
Baseline normal GPU memory allocation sizes per workload and alert on deviations

How to Mitigate CVE-2026-16280

Immediate Actions Required

Apply the vendor-supplied patch from Imagination Technologies as soon as available for your platform
Inventory all systems running Imagination Technologies GPU drivers and prioritize patching
Restrict access to GPU compute interfaces on multi-tenant systems until the patch is deployed

Patch Information

Imagination Technologies has published guidance in the Imagination Technologies Security Advisory. Downstream vendors that integrate the affected driver (system-on-chip integrators, mobile OEMs, and Linux distributions) must ship the fix through their update channels. Verify driver versions on affected devices match or exceed the patched release referenced in the advisory.

Workarounds

Where patching is not immediately feasible, restrict GPU driver access to trusted users through group permissions on the driver device node
Limit maximum sparse PMR allocation sizes below the 4 GB threshold via driver configuration where supported
Disable GPU compute functionality on systems that do not require it

bash

# Configuration example: restrict GPU device node access to a trusted group
sudo chown root:gpu-trusted /dev/pvr_sync
sudo chmod 660 /dev/pvr_sync