GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux mm: file-backed THP split reads folio state after dropping the last reference

CVSS 7.8CVE-2026-53189Kernel, userspace & hypervisorcurated

Impact

When a file or shmem PMD mapping is split, the kernel updates the RSS counter after releasing the folio reference. If that release frees the folio, mm_counter_file() then reads folio_test_swapbacked() out of freed memory. Any local process that can map a file or shmem huge page and force a PMD split exercises this path, and THP on shmem is the normal configuration on GPU nodes that use large hugepage-backed staging buffers or /dev/shm for inter-process tensor handoff. The record gives no exploitation detail beyond the read of freed state; treat it as kernel memory corruption of unproven reliability rather than a known escape. Scored 7.8 local by NVD.

Who can reach it

Local unprivileged user or any container able to create file/shmem-backed transparent huge pages and trigger a PMD split. No network path.

What to do

Update to a stable kernel that moves the counter update before folio_put(); the fix is a one-line reordering in core mm, so distro kernels will carry it in a routine update. Core mm cannot be mitigated by configuration short of disabling file/shmem THP, which costs performance on GPU nodes. Rolling it out means drain and reboot each node.

References

Related entries

All Kernel, userspace & hypervisor entries

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.