AMD EPYC IOMMU: reads past the Device Table let a malicious guest corrupt hypervisor pages

CVSS 6.0CVE-2024-21963Kernel, userspace & hypervisorcurated

Impact

Under specific circumstances the AMD IOMMU reads past the end of the Device Table that the OS or hypervisor allocated. It treats the next physical page, which may belong to a guest, as device configuration. AMD rates the outcome as hypervisor page corruption and a possible host crash. Xen's XSA-524 says a guest might take control of a device's IOMMU configuration, modify arbitrary host memory and escalate to hypervisor privilege. The exposure depends on the platform. Hosts are at risk where the IVRS table does not cover a full PCI bus (Device Table under 2MB) and a device sits at, or can use, an initiator ID outside that range. That makes GPU passthrough hosts with multiple tenants the setup to check first. The issue affects EPYC 7001 through 9005 and 4004/8004 series parts.

Who can reach it

A malicious guest VM on an affected AMD EPYC host, on platforms where the IVRS and device layout meet the conditions above. Requires guest-level privileges, and the attack is high complexity.

What to do

For Xen, apply the XSA-524 patch (xen-unstable and 4.17.x and the stable branches) and reboot each host into the patched hypervisor, which means migrating or stopping its guests. Xen states there are no mitigations. For KVM and other hypervisors, and for any firmware component, check AMD-SB-3060 and AMD-SB-4018 for what applies. The record does not name a firmware or kernel fix version.

References

https://nvd.nist.gov/vuln/detail/CVE-2024-21963 https://www.amd.com/en/resources/product-security/bulletin/AMD-SB-3060.html https://www.amd.com/en/resources/product-security/bulletin/AMD-SB-4018.html http://xenbits.xen.org/xsa/advisory-524.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.