Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB
CVE-2026-98002 Overview
CVE-2026-98002 is a Linux kernel vulnerability in the AMD Input-Output Memory Management Unit (IOMMU) driver. The flaw resides in amd_iommu_alloc_domain_nested(), which performs an ineffective error check after calling amd_iommu_pdom_id_alloc(). The allocation function returns a signed int, but the caller stores the result in a u32 field (gdom_info->hdom_id) before testing it. Sign information is lost during the assignment, so negative error codes such as -ENOSPC become large positive values that bypass the <= 0 validation. The nested domain is then initialized with a host domain identifier that was never actually allocated.
Critical Impact
A local attacker with low privileges can trigger nested IOMMU domain setup using an unallocated host domain ID, leading to kernel memory corruption and potential privilege escalation on AMD systems.
Affected Products
- Linux kernel versions containing the AMD IOMMU nested domain allocation code
- Systems using AMD processors with IOMMU virtualization enabled
- Virtualization hosts leveraging nested paging on AMD hardware
Discovery Timeline
- 2026-09-25 - CVE-2026-98002 published to NVD
- 2026-09-25 - Last updated in NVD database
Technical Details for CVE-2026-98002
Vulnerability Analysis
The defect is a type-handling flaw in the AMD IOMMU driver's nested domain allocation path. The helper amd_iommu_pdom_id_alloc() returns an int, encoding either a positive domain ID on success or a negative errno value propagated from ida_alloc_range() when the ID space is exhausted or memory allocation fails. The caller amd_iommu_alloc_domain_nested() assigns this signed return value directly into gdom_info->hdom_id, declared as an unsigned 32-bit integer, and only then checks if (gdom_info->hdom_id <= 0).
Because the sign is discarded during the implicit conversion, -ENOSPC is stored as 0xFFFFFFE4, which is a large positive unsigned value. The subsequent <= 0 comparison never evaluates true, so the error path is skipped. The driver proceeds to construct a nested domain referencing a host domain ID that was never allocated by the IDA (ID Allocator) subsystem.
Root Cause
The root cause is a signed-to-unsigned conversion that destroys error semantics before validation. The fix retains the return value in an int variable, validates it there, and only writes to the u32 field once the value is confirmed to be a valid domain ID, matching the pattern already used by other amd_iommu_pdom_id_alloc() callers.
Attack Vector
Exploitation requires local access with low privileges on an AMD system where IOMMU nested domains can be allocated. An attacker who can trigger nested domain allocations while the IOMMU domain ID space is exhausted causes the driver to operate on an invalid host domain reference. This can corrupt kernel data structures tracking IOMMU protection domains and compromise the confidentiality, integrity, and availability of the host and attached guests. The scope changes because the AMD IOMMU driver manages resources shared across virtualization boundaries.
No verified public exploit code is available. See the upstream patches at kernel.org commit 80c231f5 and kernel.org commit fa5c0827 for the authoritative fix.
Detection Methods for CVE-2026-98002
Indicators of Compromise
- Kernel warnings or oops messages referencing amd_iommu, pdom_id, or nested domain allocation failures under memory pressure.
- Unexpected IOMMU protection domain entries in /sys/kernel/debug/iommu/ not corresponding to known device assignments.
- Workloads repeatedly invoking VFIO or guest device-passthrough interfaces to exhaust IOMMU domain IDs.
Detection Strategies
- Compare the running kernel version against the fixed commits 80c231f5577229d1e90a1e312b3095f9502f8625 and fa5c0827f0b7bac6d0a188f10118151769ae68fd to identify unpatched hosts.
- Audit kernel ring buffer logs (dmesg) for IOMMU subsystem errors coinciding with VFIO device assignment or virtual machine startup activity.
- Monitor for unprivileged processes invoking IOMMU-related ioctl calls at unusual rates on AMD hosts.
Monitoring Recommendations
- Centralize dmesg and kernel audit logs from AMD-based virtualization hosts for anomaly detection.
- Track ida_alloc_range exhaustion patterns and kernel memory allocation failures that could precede exploitation attempts.
- Alert on new kernel modules loading or on crash dumps generated by the IOMMU driver.
How to Mitigate CVE-2026-98002
Immediate Actions Required
- Apply the upstream kernel patches referenced in commits 80c231f5 and fa5c0827 or update to a stable kernel release that includes both fixes.
- Inventory AMD-based hypervisors and bare-metal systems using IOMMU passthrough and prioritize them for patching.
- Restrict local shell access and VFIO device-assignment capabilities to trusted administrators until patched.
Patch Information
The fix keeps the return value of amd_iommu_pdom_id_alloc() in a signed int, tests it for a negative error before use, and assigns it to gdom_info->hdom_id only after validation. Patches are available in the stable Linux kernel tree at kernel.org commit 80c231f5 and kernel.org commit fa5c0827. Distribution vendors typically backport these changes; consult your Linux distribution's security advisories for packaged updates.
Workarounds
- Disable AMD IOMMU where device passthrough is not required by booting with amd_iommu=off or omitting iommu=pt from kernel parameters, understanding the loss of DMA protection.
- Avoid granting unprivileged users access to /dev/vfio/* and related interfaces that can drive nested domain allocation.
- Limit the number of simultaneous nested IOMMU domains to reduce the likelihood of exhausting the ID space.
# Verify running kernel version and confirm the fix is present
uname -r
# Check for AMD IOMMU initialization and current domain usage
dmesg | grep -iE 'AMD-Vi|iommu'
# On systems that do not require device passthrough, disable AMD IOMMU
# by editing /etc/default/grub and adding to GRUB_CMDLINE_LINUX:
# amd_iommu=off
# Then regenerate the bootloader config:
sudo grub2-mkconfig -o /boot/grub2/grub.cfg
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB

