CVE-2023-34326 is a flaw in how Xen invalidates IOMMU caches on AMD systems, tracked as XSA-442 and scored 7.8 High. A guest with a passed through device could reach memory it does not own. The advisory provides patches and says not to pass devices to guests.
What's affected
Per XSA-442, all Xen versions that support PCI passthrough are affected, but only x86 AMD systems with IOMMU hardware, and only x86 guests that have physical devices passed through. The AMD-Vi specification's cache invalidation guidance is incorrect on some hardware, so devices can use stale DMA mappings when some fields of the device table entry change without an IOMMU TLB flush.
Stale mappings may point to memory the guest does not own. Impact listed: privilege escalation, host-wide denial of service and information leaks. Patches: xsa442.patch for unstable, xsa442-4.17.patch for 4.17 and 4.16, xsa442-4.15.patch for 4.15. The patches may not apply cleanly to the latest release tarball, so update to the tip of your stable branch first.
How to check if you're exposed
Check three things. First, the CPU: this affects AMD hosts with an IOMMU, so confirm the vendor with lscpu. Second, the Xen version from xl info. Third, whether any guest has a physical device passed through. Look for pci entries in guest configuration:
grep -r "^pci" /etc/xen/
If you find no passthrough on AMD hosts, the advisory's own mitigation, not passing devices to guests, already describes your state.
What to do now
- Apply the XSA-442 patch for your branch, or install a vendor build that contains it.
- Update to the tip of your stable branch first if the patch does not apply cleanly.
- Where patching must wait, remove PCI passthrough from guests on AMD hosts.
- Review which guests really need device passthrough, such as GPUs or network cards.
- Record AMD hosts and passthrough use in your asset list.
Why it matters
This one is a hardware documentation problem. The specification said one thing about when caches need flushing, and some hardware needs more. Software that follows the guidance exactly is still wrong.
IOMMUs exist to confine devices to the memory they should reach. When mappings go stale, that confinement fails quietly. The narrow scope, AMD hosts with passthrough, limits exposure, though cloud and lab hosts that hand GPUs to guests fit it well.
The weakness classes involved are CWE-672 (Operation on a Resource after Expiration or Release). The record does not flag this CVE as listed in CISA's Known Exploited Vulnerabilities catalog, and none of the advisories I read says it is exploited in the wild. That can change, so treat the status as a snapshot.
The record scores this CVE 7.8 High. The Xen advisory gives no CVSS score and names no fixed release, only patches.
Quick check
Does this affect Intel hosts?
The advisory says only x86 AMD systems with IOMMU hardware are affected.
What is the simplest mitigation?
Do not pass physical devices through to guests, per the advisory.
Sources
Last reviewed: 19 October 2026.
