{"id":"CVE-2026-98156","published":"2026-09-25T11:17:47.280","lastModified":"2026-09-25T15:18:07.240","description":"In the Linux kernel, the following vulnerability has been resolved:\n\ndrm/virtio: use the DMA API for resource backing on Xen\n\nOn a Xen PV domain page addresses bear no relation to the real machine\naddresses the host would have to use to reach it.\nvirtio_ring.c handles this correctly, vring_use_map_api() returns true\nfor any xen_domain() regardless of VIRTIO_F_ACCESS_PLATFORM.\n\nvirtio-gpu makes the same decision independently, but its copy\nlooks only at the feature bit:\n\n\tbool use_dma_api = !virtio_has_dma_quirk(vgdev->vdev);\n\nQEMU does not set iommu_platform on virtio-vga by default, so\nVIRTIO_F_ACCESS_PLATFORM is not negotiated, use_dma_api is false, and\nvirtio_gpu_object_shmem_init() describes the framebuffer's backing pages\nto the host with sg_phys().  Those are guest-physical addresses. In a PV\ndomain they resolve, on the host side, to pages belonging to some other\ndomain, so the host scans out unrelated memory.\n\nMove the decision into virtio_gpu_use_dma_api() and give it the\nxen_domain() check, like vring_use_map_api() has. This\nadditionally enables the dma_sync_sgtable_for_device() calls in\nvirtgpu_vq.c, which are required for correctness whenever swiotlb\nis in play.\n\nReproduced with a Xen 4.21 PV dom0 nested inside QEMU 8.2 with\nvirtio-vga, on both a distro 6.8 kernel and 6.18 LTS. A PVH dom0\nworks fine and doesn't need this fix because it is identity-mapped,\nonly PV dom0s are affected.","cvssScore":7.8,"cvssSeverity":"HIGH","cvssVector":"CVSS:3.1/AV:L/AC:H/PR:L/UI:N/S:C/C:H/I:H/A:H","cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/455184dbe9652d01470d3c5cf5edd09a7a673a99","tags":[]},{"url":"https://git.kernel.org/stable/c/6a736d2f9d0c6e6217fe7532bc4c50ceca71db78","tags":[]},{"url":"https://git.kernel.org/stable/c/93c557e947ef0040004b1ac6a1c1265d79a957e1","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}