Linux Kernel EUVD-2026-86885

Requires a Linux system running an affected kernel (6.5 through the fixed 6.12.111/6.18.53/7.2.7/7.3-rc3) that actually has Rockchip VPU981-class AV1 decoding hardware and an exposed stateless AV1 V4L2 decoder device node (/dev/videoN). A local, low-privileged attacker with access to that device node must submit a crafted AV1 frame whose V4L2 tile controls declare tile_cols * tile_rows exceeding either the count of submitted tile-group entries or AV1_MAX_TILES. Limiting factors: no network vector (AV:L), the specific Rockchip media hardware and device-node permissions must be present, and no user interaction by others is needed - but the hardware dependency sharply narrows the exposed population versus a generic kernel flaw.

This is a driver-specific bounds-check flaw in the Linux kernel Rockchip VPU981 stateless AV1 video decoder (media/verisilicon), and it is the driver-level twin of the generic v4l2-ctrls tile-validation issue CVE-2026-97575. The provided CVSS vector (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H = 7.8 HIGH) is internally consistent about the LOCAL attack vector and low-privilege requirement, but the C:H/I:H/A:H triple is overstated for what the code actually does: the primary defect is an out-of-bounds READ of the tile-group-entry array plus hardware being programmed for more tiles than the descriptor buffer holds, which is crash- and memory-corruption-dominant rather than a clean full-confidentiality/integrity compromise. There is NO EPSS score, NO CISA KEV listing, and NO public PoC provided, so all real-world-exploitation signals are absent or point to low urgency. Additionally, exploitation is doubly gated by hardware: it requires a system with an actual Rockchip VPU981 (RK3588-class) AV1 decoder AND local access to its /dev/videoN interface - a far smaller footprint than 'Linux kernel' implies. Net: a genuine bug worth patching on affected Rockchip media hardware, but not a broad-priority remote threat; the 7.8 rating overstates real-world risk. Note also a data-quality problem - the input carries a 'Mediatek' tag, but MediaTek is only referenced in the description as the driver whose bound this fix mirrors; the affected product is Rockchip/Verisilicon.

A local user with access to the Rockchip AV1 decoder V4L2 device submits a crafted AV1 frame whose V4L2 controls claim more tiles (tile_cols * tile_rows) than the number of tile-group entries actually submitted, or more than AV1_MAX_TILES. The decoder's set_tile_info() then indexes past the dynamically-sized entry array (out-of-bounds read) and programs the hardware for more tiles than the descriptor buffer holds, causing memory corruption or a kernel crash. No public exploit code is identified at time of analysis, and the attack requires local access to the specific Rockchip VPU hardware.