CVE-2026-98156: drm/virtio: use the DMA API for resource backing on Xen
drm/virtio: use the DMA API for resource backing on Xen
Affected Software
Event History
Frequently Asked Questions
Which systems are exposed to this issue?
The issue affects Linux kernel systems using virtio-gpu in a Xen PV domain. It is specifically relevant when the guest's framebuffer backing pages are provided to the host using guest-physical addresses rather than DMA-mapped addresses.
Is the default QEMU virtio-vga configuration affected?
Yes. QEMU does not enable iommu_platform on virtio-vga by default, so VIRTIO_F_ACCESS_PLATFORM is not negotiated; the prior virtio-gpu logic could therefore avoid the DMA API even in a Xen PV domain.
What is the observable impact on an affected guest?
The host can resolve the supplied guest-physical framebuffer addresses to pages belonging to another domain and scan out unrelated memory. This can result in display output containing memory that does not belong to the affected guest.
What changes in the resolved kernel behavior?
The DMA API decision now includes a Xen PV domain check, matching virtio ring handling. This causes framebuffer backing resources to use DMA mappings and enables required DMA synchronization when swiotlb is in use.