AMD EPYC IOMMU: improper exception handling sends ATC invalidations to devices without ATC support

CVSS 7.0CVE-2024-21954Firmware, BMC & network fabriccurated

Impact

Improper exception handling in the IOMMU of EPYC 7003 and 9004 processors (including the Embedded parts) can cause an Address Translation Cache invalidate command to be sent to a PCIe device that does not support ATC. The result is denial of service, and the CVSS vector marks the availability impact as reaching beyond the vulnerable component. On GPU hosts, the IOMMU sits between every passthrough GPU, NIC and NVMe device and the rest of the system. A fault there can disrupt the whole node and every tenant on it, not just one device. The record gives no detail on the trigger conditions.

Who can reach it

Low-privileged access is required. The CVSS vector rates attack complexity as high and lists attack prerequisites. The record does not say which actor, for example a guest with a passthrough device or a host user, can trigger the invalidation.

What to do

See AMD-SB-3052 for the platform firmware that carries the fix on EPYC 7003 and 9004. The record does not list a version. Platform firmware updates on these systems normally mean a BIOS flash and a reboot of each node, so plan to drain the node first.

References

https://nvd.nist.gov/vuln/detail/CVE-2024-21954 https://www.amd.com/en/resources/product-security/bulletin/AMD-SB-3052.html

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.