Intel processors, including Core and Xeon products, are susceptible to attacks described as Snoop-assisted L1 data sampling, a security investigator has discovered.
Described in a deep-dive report published to warn software developers, Intel said the Snoop attacks were first brought to light by a software engineer.
The attacks allow bad actors to insert malicious code after a change in the L1D cache, at which point the CPU updates all cache levels in a process known as bus snooping. The L1 data sampling technique could then trigger errors to leak data from the CPU’s inner memory.
Cybercriminal Activity at the Cache Level
On a more technical level, the Snoop attacks are taking advantage of cache coherence, which ensures that the L1D cache is using data that is consistent with whatever has been stored across L1, L2 and L3 caches, along with the RAM. This is common in many of the multi-core architectures that run within server environments today, the report noted.
Companies are able to set up virtualized IT infrastructure and cloud computing environments, in part, because of the multi-core capabilities of modern chips made by hardware developers like Intel. If malware that infects a CPU via Snoop attacks leaks data from other cores, however, a single attack could have significant consequences.
On the other hand, it’s important to recognize that Snoop attacks are largely theoretical at this point. The Intel report said that creating the real-world conditions necessary to carry out the technique would be difficult.
Unlike some of the other vulnerabilities affecting Intel processors, such as Spectre and Meltdown, Intel said a Snoop attack would be unlikely to steal much data that cybercriminals could use.
Apply the Foreshadow Patches
Fortunately, Intel said those who might be affected by the vulnerability would be able to protect themselves by applying patches released two years ago, dubbed Foreshadow (L1TF). Turning off a feature in Intel CPUs known as Transactional Synchronization Extensions (TSX) will also make Snoop attacks harder to pull off, the company said.