New TONTOU Attack Defeats Spectre v2 Defenses on Intel and AMD CPUs
Security researchers from MIT have uncovered a new attack called TONTOU that bypasses Spectre v2 protections on modern Intel and AMD processors. The exploit targets a critical timing gap in neutralization-based defenses, where the CPU cleanses the branch predictor state but leaves a brief window before that state is actually used. By exploiting this window, attackers can re-poison the processor's internal state and gain access to sensitive information.
The researchers demonstrate that unprivileged programs can schedule timer interrupts to trigger during kernel execution, forcing the system to redirect to an interrupt handler. This handler then poisons microarchitectural states within the vulnerable post-neutralization window. The practical impact is severe, as attackers can extract secrets such as Linux password hashes without requiring any special system privileges or elevated access.
The findings affect all modern processors relying on eIBRS from Intel and Safe RET from AMD, which are the current industry-standard mitigations against Spectre v2 branch target injection attacks. While the researchers have responsibly disclosed their work, the discovery highlights that speculative execution vulnerabilities remain a persistent challenge for chipmakers. Organizations are advised to monitor for vendor patches and consider additional isolation strategies to protect sensitive kernel data.