Security researchers from Google have issued a report regarding vulnerabilities (cve-2024-56161) found in AMD processors. These vulnerabilities impact the microcode loader, allowing for the bypassing of the digital signature verification mechanism during microcode updates. By loading modified microcode, attackers can compromise the AMD SEV (Secure Encrypted Virtualization) mechanism used in virtual systems to protect virtual machines from interference by the hypervisor or host system administrators.
The vulnerabilities stem from the use of insecure hash functions in the code that performs digital signature checks after loading microcode into the CPU. To exploit the vulnerabilities, an attacker must possess administrator rights on the local system, enabling code execution at Ring0 level outside the virtual machine.
Attackers can infiltrate guest systems secured using AMD SEV (Secure Encrypted Virtualization) and SEV-SNP (Secure Nested Paging) extensions, which ensure memory integrity in virtual machines, isolate processor registers, and facilitate secure operations with nested memory page tables. These mechanisms were designed to prevent data center and cloud provider staff from altering or examining memory contents of protected guest systems, as well as manipulating calculations.
To demonstrate the severity of the vulnerability, researchers have proposed modified microcode that alters the logic of RDRAND instructions—a source of entropy in key generation, cryptographic operations, and random identifier creation—certified with a digital signature. The modified microcode instructs RDRAND instructions to return a fixed value of 4, instead of a random sequence, by reserving the CF flag (Carry Flag), marking the output as erroneous. Further details and tools for implementing the modified microcode are set to be released on March 5 to allow users time to apply corrections. An attack demonstration was conducted on servers equipped with AMD EPYC 7B13 (Milan) and AMD Ryzen 9 7940HS (Phoenix) processors.