Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB
CVE-2026-98518 Overview
CVE-2026-98518 is a Linux kernel vulnerability in the nouveau DRM driver's device memory (dmem) subsystem. The flaw stems from a size mismatch between the registered device-private memory region and the VRAM buffer object that backs it. A prior commit expanded the device-private region registered by nouveau_dmem_chunk_alloc() from DMEM_CHUNK_SIZE to DMEM_CHUNK_SIZE * NR_CHUNKS to support Transparent Huge Pages (THP) for GPU memory migration. The backing VRAM buffer object was not resized. As a result, address translation for pages beyond the first chunk resolves to VRAM outside the allocated buffer object.
Critical Impact
Memory accesses to migrated GPU pages beyond the first chunk resolve to VRAM outside the backing buffer object, leading to out-of-bounds VRAM access on affected Nouveau-enabled systems.
Affected Products
- Linux kernel versions containing commit c32287471077 ("gpu/drm/nouveau: enable THP support for GPU memory migration")
- Linux distributions shipping the affected drm/nouveau/dmem subsystem
- Systems using the Nouveau DRM driver for NVIDIA GPU memory migration
Discovery Timeline
- 2026-10-18 - CVE-2026-98518 published to NVD
- 2026-10-19 - Last updated in NVD database
Technical Details for CVE-2026-98518
Vulnerability Analysis
The vulnerability resides in the Nouveau DRM driver's device memory management code. nouveau_dmem_chunk_alloc() registers a device-private memory region used to migrate pages between system memory and GPU VRAM. Commit c32287471077 enlarged this registered region to DMEM_CHUNK_SIZE * NR_CHUNKS to enable Transparent Huge Page support. The commit left the underlying VRAM buffer object (chunk->bo) sized at DMEM_CHUNK_SIZE.
The address resolution function nouveau_dmem_page_addr() returns chunk->bo->offset plus the page's offset within the registered region. When the page offset exceeds DMEM_CHUNK_SIZE, the returned VRAM address falls outside the buffer object's bounds. This produces out-of-bounds reads and writes to GPU VRAM during page migration operations.
Root Cause
The root cause is a size mismatch between the registered device-private memory range and its backing buffer object. The registered region covers NR_CHUNKS worth of address space, but the VRAM allocation covers only a single chunk. Any page past the first chunk translates to VRAM addresses the driver never allocated for this purpose.
Attack Vector
Triggering the condition requires GPU memory migration workloads that span more than one chunk on systems running the Nouveau driver with the THP-enabled code path. Local workloads performing page migration to device memory can cause the driver to compute VRAM offsets outside the backing buffer object, resulting in memory corruption or data integrity issues within GPU memory. The fix sizes the buffer object to match the region it backs.
Detection Methods for CVE-2026-98518
Indicators of Compromise
- Kernel log messages referencing nouveau_dmem, VRAM allocation faults, or GPU memory migration errors
- Unexpected GPU resets, hangs, or page fault entries from the nouveau driver during HMM or device-private memory operations
- Corruption in compute or graphics workloads using GPU memory migration on Nouveau-driven NVIDIA hardware
Detection Strategies
- Inventory kernel versions across Linux hosts and flag kernels that include commit c32287471077 but lack the fix commits 06c69c64686a or 64ca4cdd1031
- Monitor dmesg and journald for nouveau driver warnings tied to device-private memory, buffer object offsets, or VRAM access
- Correlate GPU workload failures with kernel version data to identify systems running vulnerable builds
Monitoring Recommendations
- Collect kernel version telemetry from Linux endpoints and compare against patched stable kernel releases
- Alert on repeated nouveau driver faults or GPU subsystem errors on workstations and compute nodes
- Track vendor advisories from Linux distributions for backported fixes to the drm/nouveau/dmem subsystem
How to Mitigate CVE-2026-98518
Immediate Actions Required
- Identify Linux systems running the Nouveau driver with GPU memory migration enabled
- Apply the upstream kernel fixes referenced by commits 06c69c64686a and 64ca4cdd1031 or the equivalent distribution update
- Prioritize patching on hosts that perform GPU compute workloads using device-private memory migration
Patch Information
The fix sizes the VRAM buffer object to match the full registered device-private region. Patches are available in the Linux stable tree via the Kernel Git Commit 06c69c64686a and Kernel Git Commit 64ca4cdd1031. Linux distributions typically backport such fixes; update to the latest vendor-supplied kernel package.
Workarounds
- Where patching is not immediately possible, avoid GPU memory migration workloads on Nouveau-driven systems
- Consider switching to the proprietary NVIDIA driver on affected hosts if operational requirements permit
- Unload the nouveau module on systems that do not require GPU acceleration via this driver
# Verify running kernel version and nouveau module status
uname -r
lsmod | grep nouveau
# Optionally blacklist nouveau until a patched kernel is deployed
echo 'blacklist nouveau' | sudo tee /etc/modprobe.d/blacklist-nouveau.conf
sudo update-initramfs -u
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB

