New Interrupt Injection Attack Bypasses Spectre v2 Defenses on Intel and AMD CPUs

Researchers from MIT CSAIL have unveiled a novel attack technique called Interrupt Injection that can bypass Spectre v2 mitigations on Intel and AMD processors. The attack exploits a timing gap between when a processor sanitizes its branch predictor and when the kernel uses it, allowing an unprivileged local attacker to re-poison the predictor and leak sensitive kernel memory.

On an AMD Zen 2 system running Linux 6.14 with all default mitigations enabled, the exploit achieved a data leak rate of 5.47 bytes per second with 91.97% accuracy, successfully reading /etc/shadow in five out of ten attempts. The technique requires no special privileges, only local code execution, making shared systems particularly vulnerable.

The researchers disclosed the vulnerability to AMD and Intel on February 5. AMD has acknowledged the issue in bulletin AMD-SB-7061, titled “Safe RET Interrupt Vulnerability,” affecting Zen 1 through Zen 4 processors. AMD states that a kernel patch has been shipped and will arrive via normal OS updates. Intel, however, does not consider a mitigation necessary, though it paid a discretionary bug bounty bonus.

The attack class is named TONTOU (Time-of-Neutralization to Time-of-Use), drawing parallels to TOCTOU races in software. The researchers demonstrated the attack on multiple CPU generations, with varying success rates. The work was presented at Black Hat USA and is scheduled for USENIX Security.