TLBleed Side-Channel CPU Attack Detailed at Black Hat

LAS VEGAS—This has been a busy year for researchers disclosing new attacks against Intel chips. On Aug.10 at Black Hat USA here, security researcher Ben Gras from VU University detailed the TLBleed side-channel attack that potentially exposes Intel CPUs to risk.

The initial research on TLBleed was disclosed on June 22, and Gras provided full details in his Black Hat session. The attack abuses the Translation Lookaside Buffer (TLB) capability in modern CPUs that provides a memory cache used to help accelerate performance. According to Gras, the attack works on Intel CPUs with hyperthreading enabled.

"With TLBleed we show that even with cache protections in place, we can still leak secrets from a cached process," Gras said in a press conference at Black Hat following his session.

With TLB, is it possible for multiple process threads to have concurrent access. As such, with the TLBleed attack, Gras found a weakness that might enable an attacker to gain access to the shared memory as well as access to information on another process, bypassing existing cache protection features. Gras said Intel has downplayed the security flaw, telling him that there is already guidance for developers on how to secure TLB, by writing the correct cryptographic algorithms that don't leave data traces in cache.

In response to a question from eWEEK, Gras noted that the attack is unrelated to the Meltdown and Spectre CPU vulnerabilities that were disclosed on Jan. 3.

"Spectre and Meltdown rely on CPU artifacts like out-of-order execution and speculative execution," he said. "TLBleed observes shared caches."

While TLBleed is different, it is still a form of side-channel attack, which has been area of active interest for VU University and other researchers throughout 2018. Gras said he expects there to be continuing disclosures in 2018 and beyond about more side-channel and Spectre variant attacks. 

"I see this [Spectre] as being like the first buffer overflow, which has been around since the 1990s, but even today we still find new buffer overflow flaws," he said.  

Watch the full video with VU University Security Researcher Ben Gras above.

Sean Michael Kerner is a senior editor at eWEEK and Follow him on Twitter @TechJournalist.