CVE-2024-56538

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 vulnerability is reached through ioctl and mmap on the local DRM primary node /dev/dri/card0 of the zynqmp-dpsub device. There is no network or remote-peer input path into this driver.

AC:LAttack complexity, low

This is not a narrow race — because drm_dev_unregister() never sets dev->unplugged, the guard state is broken permanently, so every subsequent ioctl/mmap from an already-open fd deterministically enters the torn-down driver. An attacker merely holds an fd open and loops ioctls, and the freed DMA channels, unmapped iomem, and destroyed mutex are hit reliably.

PR:LPrivileges required, low

The attacker only needs an open file descriptor on the DRM primary node, which unprivileged local users of a graphical session obtain via logind ACLs or video-group membership on these embedded/kiosk systems. The reachable dumb-buffer ioctls carry flags 0, requiring neither DRM master nor DRM_AUTH.

UI:NUser interaction, none

The attacker's process simply keeps its fd open and polls; no victim action is needed at exploitation time. On ZynqMP FPGA MPSoCs the driver teardown occurs through routine platform events such as FPGA-region/device-tree-overlay reconfiguration and module reload rather than a deliberate victim action.

S:UScope, unchanged

The corruption is confined to kernel memory within the same security authority as the kernel itself. No VM, IOMMU, or sandbox boundary is crossed.

C:HConfidentiality impact, high

The missing unmap_mapping_range() leaves userspace with live mappings of GEM DMA buffers whose backing reserved-memory pool was released by of_reserved_mem_device_release(), giving a direct read primitive into kernel memory that can be recycled for other allocations. The use-after-free on released DMA channels and driver state further enables disclosure of freed kernel object contents.

I:HIntegrity impact, high

Those same stale userspace mappings are writable, yielding a direct write-after-free into memory the kernel will reallocate, and the driver continues submitting DMA descriptors on channels already handed back via dma_release_channel(). This combination provides heap-grooming and control-flow-hijack primitives typical of an exploitable use-after-free.

A:HAvailability impact, high

Post-removal ioctls write to devres-unmapped iomem with the APB clock gated and lock a mutex_destroy()'d dp->lock, producing an oops or an AXI bus hang that wedges the SoC. Any use-after-free of this kind reliably crashes the kernel even when not weaponized further.