Executive brief
A memory leak was identified in the AMD GPU driver within the Linux kernel. This occurs during the initialization of Reliability, Availability, and Serviceability (RAS) features if a specific sub-component fails to start. While primarily a stability issue, repeated occurrences could eventually exhaust system memory, potentially leading to a system crash or performance degradation.
Technical details
A memory leak exists in the amdgpu_ras_init() function within drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c. When the call to amdgpu_nbio_ras_sw_init() fails, the function returns an error code immediately without jumping to the cleanup label (release_con). This leaves the 'con' structure allocated in memory with no remaining references. An attacker or a series of hardware initialization failures could trigger this path to exhaust kernel memory. The issue has been resolved by ensuring the function jumps to the appropriate cleanup label upon failure.
Affected products
- Linux Linux Kernel Versions including amdgpu driver with RAS support
Timeline
- 2026-01-29: other: Patch authored
- 2026-02-05: patched: Mainline patch committed
- 2026-05-27: disclosed: CVE published
References
- https://git.kernel.org/stable/c/2fef8c2ac67e7c1b0409d23653300b134c63e54c
- https://git.kernel.org/stable/c/3f43e7812b30d6b2e850218f9bb1dae60727fcef
- https://git.kernel.org/stable/c/c11cd77a18115d2cd3f4b6915c4a537b6042f950
- https://git.kernel.org/stable/c/ee41e5b63c8210525c936ee637a2c8d185ce873c
- https://git.kernel.org/stable/c/f8a5426652bdadd4a5cb48326d48abbdfebe8153