Why do "red processors" always output "4" when generating random numbers?
Researchers from Google have discovered a critical vulnerability in AMD processors that allows unofficial microcode to be loaded and altered. Thanks to the vulnerability, the specialists were able to force the processors to always return the number 4 when generating random numbers. This compromises AMD's defense mechanisms, including SEV-SNP, a secure virtualization system, and the root of trust.
Microcode is the built-in programs that control the processor at a low level. AMD protects its microcode updates with a cryptographic signature to prevent substitution. However, Google specialists have found a way to bypass the protection using a vulnerability in the hash algorithm for verifying the signature. Their method works on all Zen-based processors, including Ryzen and Epyc.
To prove the possibility of modifying the microcode, the researchers released a demonstration patch for server Epyc (Milan) and desktop Ryzen 9 (Phoenix) processors, forcing the RDRAND instruction to always return 4. Potential attacks can use similar techniques to weaken cryptographic protection.
Despite the severity of the issue, loading the modified microcode requires kernel-level (ring-0) access, which limits the attack to system administrators or advanced malware. However, in cloud environments with trusted virtualization, microcode substitution on the host can undermine the security of guest machines.
SEV-SNP, AMD's mechanism for protecting virtualized workloads, can also be compromised by this method, allowing attackers to spoof compute processes and memory access.
AMD has acknowledged the vulnerability and has already released a fix in the form of updated microcode for server and embedded processors. Updates will be released later to protect desktop Ryzen and Threadripper. A BIOS update with the fixed microcode will allow systems to automatically verify the authenticity of downloaded updates and block unofficial versions.
The vulnerability has been identified as CVE-2024-56161 and has a CVSS score of 7.2. AMD recommends that all users update their motherboard firmware to prevent potential attacks. Google promises to publish additional details about the vulnerability on March 5, 2025.
Researchers from Google have discovered a critical vulnerability in AMD processors that allows unofficial microcode to be loaded and altered. Thanks to the vulnerability, the specialists were able to force the processors to always return the number 4 when generating random numbers. This compromises AMD's defense mechanisms, including SEV-SNP, a secure virtualization system, and the root of trust.
Microcode is the built-in programs that control the processor at a low level. AMD protects its microcode updates with a cryptographic signature to prevent substitution. However, Google specialists have found a way to bypass the protection using a vulnerability in the hash algorithm for verifying the signature. Their method works on all Zen-based processors, including Ryzen and Epyc.
To prove the possibility of modifying the microcode, the researchers released a demonstration patch for server Epyc (Milan) and desktop Ryzen 9 (Phoenix) processors, forcing the RDRAND instruction to always return 4. Potential attacks can use similar techniques to weaken cryptographic protection.
Despite the severity of the issue, loading the modified microcode requires kernel-level (ring-0) access, which limits the attack to system administrators or advanced malware. However, in cloud environments with trusted virtualization, microcode substitution on the host can undermine the security of guest machines.
SEV-SNP, AMD's mechanism for protecting virtualized workloads, can also be compromised by this method, allowing attackers to spoof compute processes and memory access.
AMD has acknowledged the vulnerability and has already released a fix in the form of updated microcode for server and embedded processors. Updates will be released later to protect desktop Ryzen and Threadripper. A BIOS update with the fixed microcode will allow systems to automatically verify the authenticity of downloaded updates and block unofficial versions.
The vulnerability has been identified as CVE-2024-56161 and has a CVSS score of 7.2. AMD recommends that all users update their motherboard firmware to prevent potential attacks. Google promises to publish additional details about the vulnerability on March 5, 2025.
