The Spectre and Meltdown chip vulnerabilities could have been resolved much earlier had chip makers taken reports from academic researchers more seriously, says one researcher who helped unveiled the hardware bug.
Daniel Gruss, a researcher at Graz University of Technology, hasn’t had a break since Meltdown and Spectre came to light. Chip vulnerabilities are multiplying with increasingly complex chip designs and the emergence of new technologies such as GPUs and confidential computing.
“I think the number of bugs that we have in our systems will not get less over time,” Gruss says.
Gruss and Intel fellow Anders Fogh will reflect on past chip vulnerabilities and explore emerging threats during their Black Hat USA 2024 on Thursday, Aug. 8. The presentation Microarchitecture Vulnerabilities: Past, Present, and Future will talk about recent side-channel attack techniques as exposed by Hertzbleed, Platypus, and Zenbleed. Gruss and Fogh will also explore how academic researchers and chip makers are collaborating to counter vulnerabilities and discuss top-line mitigation and patching strategies.
Gruss, now a professor in information security, said the chip makers hadn’t been as responsive as the companies are now. His team reported the prefetch side-channel at the center of Spectre to Intel in 2016, but the chip maker dragged its feet.
“Intel could have had Spectre two years earlier than they had it… if they would just have looked at our report a bit more closely and tried it out for a longer time on different machines and then investigated, but they didn’t,” Gruss said.
But that has changed, and Intel takes every security flaw reported very seriously, Gruss said.
Communication Is Key
Intel is in lockstep with researchers, and also keeps communication lines open with rivals such as AMD and Nvidia as hardware bugs could affect multiple vendors, says Suzy Greenberg, vice president for Intel Product Assurance and Security Group.
Spectre and Meltdown used side-channel attacks to leak sensitive data that could include usernames and passwords. Hackers can conduct side-channel attacks by utilizing system functions such as frequency scaling and power consumption patterns.
Hundreds of papers on side-channel attacks have come out since the bugs were initially reported. However, no real-world break-ins based on the bugs have been reported, yet, according to Gruss and Intel. Side-channel attacks will always be there, and chip vendors won’t be able to solve the bugs, Gruss says.
“The question is … how can we keep them restricted enough so that attackers cannot exploit them for valuable information,” Gruss says.
Researchers Shift Focus to GPUs
Researchers are also shifting their attention to exploring security bugs in graphics processing units, which are chips being used to serve AI.
A team of researchers including Gruss recently published research about a side-channel attack on Nvidia’s GPUs. Nvidia last month issued 10 security alerts related to its GPU drivers and virtualization software.
“As we understand more and more about the microarchitecture on GPUs, and as they get more complex, we will also see more complex and more impactful attacks,” Gruss says.
Side-channel attacks may also increase in the realm of confidential computing, which involves creating a secure enclave within hardware to run protected applications. Top chip makers Intel and AMD offer confidential computing chips for AI applications.
“Confidential computing adds attack surface from an academic perspective … there is more to attack there than if you would be an unprivileged attacker,” Gruss says.
Privileged users can get access to interfaces, instructions and model-specific registers, which widens the attack surface.
There are a lot of new use cases and exploits that are going to start to come with AI, Intel’s Greenberg said.
“We’re really trying to encourage that community to start looking at poking there, because that’s the big unknown,” Greenberg says.
Source: www.darkreading.com