Tech

AMD contradicts: Researchers find security holes in AMD processors

Not only Intel processors have security problems: researchers at the universities of Graz (Austria) and Rennes (France) want to discovered two vulnerabilities in AMD processors. AMD CPUs have been affected since 2011 up to current CPUs from 2019, the end of the research project. AMD contradicts the researchers: there are no new security gaps, users are not at risk.

Job market

  1. Preh GmbH, Bad Neustadt (Saale)
  2. BARMER, Wuppertal, Schwäbisch Gmünd



To reduce the power consumption of their processors at the same speed, AMD introduced a way predictor for the L1D cache in 2011. The researchers were able to discover the security gaps by reverse engineering the undocumented prediction function. The first attack the researchers call Collide + Probe allows them to monitor memory access, "without knowing the physical addresses or the shared memory", it says in the paper. However, this only works if the researchers can execute code on the same logical processor core. With Load + Reload, the second attack, this also works if the code is executed on another logical processor core.

According to the researchers, the code executed on the same or a connected logical processor can also be JavaScript code on a website. With this, for example, the memory scrambling of the kernel or the browser in a sandbox can be largely eliminated. The two researchers also managed to read out data, they write.

AMD had already been informed in August 2019, but has so far not delivered a microcode update to close the processor gaps. However, AMD does not consider the gaps to be new attacks on the speculative functions of its processors: "We are aware of the paper in which the researchers claim to have found potential vulnerabilities in AMD CPUs that a malicious actor can use to manipulate a cache-related function to inadvertently read user data."

However, these are combined with known and fixed security gaps in software and speculative prediction in the processor. Therefore, the vulnerabilities are not new attacks on speculative prediction. According to AMD, classic security recommendations help against such side-channel attacks: keep software up-to-date, follow safe programming guidelines and use the latest versions of libraries.

Please activate Javascript.

Or use that Golem pure offer

and read Golem.de

  • without advertisement
  • with javascript turned off
  • with RSS full text feed