GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel arm_spe: integer overflow in PERF_IDX2OFF with AUX buffers of 2 GiB or more

CVSS 7.8CVE-2025-40081Kernel, userspace & hypervisorcurated

Impact

PERF_IDX2OFF() computed an offset without casting nr_pages to unsigned long, so an Arm Statistical Profiling Extension session with an AUX ring buffer of 2 GiB or larger overflows the calculation and the driver writes profile records at a wrong offset. This matters only on Arm64 hosts with SPE present and enabled, which in a GPU fleet means Grace and other Arm-based head and compute nodes used for profiling. Consequence is memory corruption inside the perf AUX path and a likely host crash rather than a tenant-visible escape; the record gives no evidence of a cross-tenant path. Treat it as a stability and hardening fix for nodes where profiling is actually run.

Who can reach it

Local only, and only on Arm64 hardware exposing SPE. Requires the ability to open a perf AUX session with a very large buffer, which under the default perf_event_paranoid setting means a privileged user. Unreachable on x86 nodes and on Arm nodes where SPE is not enabled.

What to do

Pick up the stable kernel commits listed in the record (the nr_pages cast). No configuration mitigation is described upstream; declining to allocate multi-gigabyte AUX buffers avoids the condition in practice. The fix ships as a kernel update, so Arm64 nodes need a drain and reboot - low urgency, fold it into the next kernel cycle.

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.