GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel arm64 SME: fixed 8-byte FFR clear corrupts memory past the SVE context struct

CVSS 7.8CVE-2023-53713Kernel, userspace & hypervisorcurated

Impact

When saving SVE state in streaming SVE mode the kernel cleared the inaccessible FFR field with an unconditional 8-byte store. With any SME vector length other than 64 bytes the store either leaves the field partly uncleared or writes past the end of a kmalloc'd context structure, which maintainers observed as kfence splats and kmalloc redzone corruption under the fp-stress selftest. That is a local kernel heap write with attacker-influenced timing, so the realistic ceiling is privilege escalation on the host, and the floor is a crash that takes the whole node with it. This only matters on arm64 hosts that actually implement SME; on x86 GPU nodes and on arm64 parts without SME the code path does not exist, so for most fleets this is a no-op. Treat it as a hygiene item for arm64 CPU hosts rather than a reason to open a window tonight.

Who can reach it

Local unprivileged user on an arm64 host whose CPU implements SME - any process that can enter streaming SVE mode and take a signal. No authentication beyond a shell or a container with normal CPU access; not remotely reachable.

What to do

Update to a stable kernel carrying the PFALSE-plus-STR-P fix (four stable branches have it, see the git.kernel.org links) and reboot the affected arm64 nodes. No configuration mitigation is documented; nodes without SME-capable CPUs are not affected and need no action.

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.