MIT boffins' TONTOU attack slips through Spectre defenses on Intel and AMD CPUs

Two MIT researchers have unveiled a new speculative execution attack that can bypass defenses against Spectre v2 on Intel and AMD CPUs. The attack, named TONTOU (Time-of-Neutralization to Time-of-Use), exploits the post-neutralization window, a period between state neutralization and usage by a victim branch. The researchers used an interrupt injection primitive to trigger interrupts during this window, allowing them to re-poison branch predictor state and manipulate speculative execution.

Practical attacks demonstrated that the TONTOU attack works on both Intel Cascade Lake Refresh and Arrow Lake processors, as well as AMD Zen 2 and Zen 4 chips. The researchers successfully broke Linux's kernel address space layout randomization (KASLR) and located sensitive data, such as the contents of /etc/shadow, containing the root password hash.

While the attack is not simple, it is practical, requiring only unprivileged code execution with timer availability on a shared kernel system. The researchers encourage further investigations into interrupt injections and TONTOU attacks, and hope to inspire more robust mitigations against Spectre-style exploits. Intel chose not to pursue additional mitigations, citing the difficulty of preventing real-world exploits, while AMD committed to addressing the issue via kernel patches.