Researchers from Vrije Universiteit Amsterdam have developed a new side-channel attack called SLAM, targeting upcoming CPUs from Intel, AMD, and Arm. SLAM exploits a memory feature designed to improve security by extracting the root password hash from kernel memory using a transient execution technique. The attack leverages unmasked gadgets, revealing sensitive data through speculative execution traces, and primarily impacts future chips that meet specific criteria, lacking strong canonicality checks in chip designs.
SLAM is based on a new class of Spectre disclosure gadgets involving pointer chasing, exploiting micro-architectural race conditions introduced by advanced hardware features like Linear Address Masking (LAM), Upper Address Ignore (UAI), and Top Byte Ignore (TBI). The attack focuses on “unmasked” gadgets using secret data as pointers, common in software, allowing attackers to observe altered cache states to infer sensitive information. The researchers identified hundreds of exploitable gadgets in the Linux kernel, demonstrating SLAM’s potential to leak arbitrary ASCII kernel data.
While Arm stated that its systems already mitigate Spectre vulnerabilities and require no additional action against SLAM, AMD pointed to current Spectre v2 mitigations for CVE-2020-12965. Intel plans to provide software guidance for future processors supporting LAM, incorporating the Linear Address Space Separation (LASS) security extension to prevent speculative address accesses. Linux engineers have developed patches to disable LAM until further guidance becomes available.