The goal behind TEEs is to ensure data on the virtual work environments (virtual machines) is secure and cannot end up manipulated or stolen.
To that end, there is a security vulnerability in AMD processors that allows attackers to penetrate virtual work environments based on the trusted computing technologies AMD SEV-ES and AMD SEV-SNP, said researchers at the CISPA Helmholtz Centre for Information Security and Graz University of Technology (TU Graz).
AMD Secure Encrypted Virtualization (SEV) is a processor extension that provides secure separation between virtual machines and the underlying software, known as the hypervisor, for managing the required resources. AMD SEV encrypts the data on the virtual machine for this purpose.
CacheWarp can undo data modifications in this working environment and fool the system into believing it has an outdated status. This is problematic, for example, if a variable determines whether a user is successfully authenticated or not. Successful authentication iusually has a “0,” which is, however, the same value with which initializes the variable. If a potential attacker enters an incorrect password, the variable is overwritten with a value not equal to “0”. However, CacheWarp can then reset this variable to its initial status when it indicated successful authentication. This allows an attacker to establish an already authenticated session.
This is possible by an unexpected interaction between CPU instructions and AMD-SEV, through which the cache can reset to its old state. Once the attacker gained access in this way, they can subsequently gain the full access rights of an administrator to the data in the virtual machine. During their tests, the researchers were able to take all the data located there, modify it and spread from the virtual machine further into the user’s infrastructure. They first bypassed the secure login and then overcame the barrier between normal user and administrator.
The researchers informed AMD about the security vulnerability in advance so the company could take the necessary measures before the research results published. AMD identified CacheWarp under the identifier CVE-2023-20592 and is providing a microcode update that fixes the vulnerability.
“Research in the field of microarchitectural attacks is fascinating because it very often reveals just how complex our modern computer systems have become,” said Andreas Kogler from the Institute of Applied Information Processing and Communications (IAIK) at TU Graz. “It’s amazing how the interplay of several factors makes it possible to extract or change data from such systems. Our work on CacheWarp shows how an attacker can make write accesses to the memory of the affected processors virtually forgotten. You can think of it like older USB sticks. If you overwrote a document there, but removed the stick before the end of the writing process, you could find parts of the old version instead of the new one the next time you plugged in and read the document.”

