GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel IOMMU DMA: allocation failure not detected in the non-blocking non-coherent path

CVSS 7.8CVE-2026-90388Kernel, userspace & hypervisorcurated

Impact

The non-blocking, non-coherent IOMMU allocation path used dma_alloc_from_pool()'s cpu_addr output to decide whether the allocation succeeded, but that function only fills cpu_addr on success and returns the page itself. On failure the code proceeded to build an IOMMU mapping from an invalid page. The record gives no attacker-controlled path and no mechanism beyond the missing check, so treat this as a failure-path robustness fix in core DMA/IOMMU code rather than a demonstrated escalation; the NVD score of 7.8 is the kernel CNA's standard local rating and is not backed by analysis in the record. It matters here because the same IOMMU DMA layer carries device passthrough and accelerator DMA, and the trigger is memory pressure in atomic context - the condition a node under a heavy training job is already in.

Who can reach it

Local: requires atomic-context non-coherent DMA allocations to fail, i.e. exhaustion of the DMA atomic pool. The record names no authenticated interface an attacker drives directly.

What to do

Pick up a stable kernel containing the listed commits and reboot the node; no configuration workaround is described, and no fixed vendor release is named. Low enough priority to fold into the next scheduled kernel roll rather than an emergency window, unless you have evidence of atomic pool exhaustion on these hosts.

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.