CVE-2026-80576

Quoted from the kernel CNA's own scoring file, word for word. A highlighted metric carries the most severe value CVSS v3.1 defines for it, which is what pushes the score up. source

AV:LAttack vector, local

The bug is reached only through the AMDGPU_CS DRM ioctl on a local /dev/dri/renderD* render node; amdgpu_cs_p2_ib() takes user-supplied ib_bytes from command-submission chunks during local GPU job setup, with no network, adjacent-radio, or physical-access path to this code.

AC:LAttack complexity, low

An attacker fully controls ib_bytes, ring selection, and backing BO mappings in each submit; triggering the flaw only requires one crafted IB chunk with length_dw above the hardware 20-bit packet limit (e.g., >0xFFFFF dwords), with no race or victim-dependent timing.

PR:LPrivileges required, low

AMDGPU_CS is registered with DRM_AUTH|DRM_RENDER_ALLOW, so any authenticated render-client process with standard local access to /dev/dri/renderD* (typical render/video group on desktops, gaming PCs, and cloud GPU VMs) can reach amdgpu_cs_ioctl() without real root or init-namespace capabilities.

UI:NUser interaction, none

Exploitation is entirely self-contained in the attacker's own DRM client (open render node, allocate/map buffers, submit CS); no separate victim action such as opening a file, mounting a filesystem, or clicking a link is required.

S:CScope, changed

On GFX/compute rings, ring_emit_ib() ORs ib->length_dw into the INDIRECT_BUFFER control dword alongside VMID and preempt/valid bits; oversized values spill into those fields, so the GPU can execute IBs under an unintended VMID and perform DMA outside the attacker's GPU virtual address space, crossing per-client GPU/IOMMU isolation (e.g., multi-tenant cloud GPU).

C:HConfidentiality impact, high

Corrupted VMID and adjacent control bits in the INDIRECT_BUFFER packet can redirect GPU fetches/execution into another client's GPU-mapped memory, providing a cross-context read primitive beyond the attacker's own BO mappings, not merely a benign driver error return.

I:HIntegrity impact, high

The same control-dword overflow can alter VALID/PREEMPT/VMID encoding so the GPU performs writes and command execution in an unintended address space, yielding cross-context memory corruption and a plausible path to arbitrary GPU-visible memory modification or further kernel compromise.

A:HAvailability impact, high

Oversized IB lengths also destabilize command submission and can hang or reset GFX/SDMA rings (as noted in the fix and related IGT oversize-packet tests), denying GPU/display availability until driver reset or reboot on affected AMD systems.