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

Two MIT researchers from the Computer Science and Artificial Intelligence Laboratory (CSAIL) have unveiled a new type of speculative execution attack called TONTOU, short for Time-of-Neutralization to Time-of-Use. This attack exploits a window of opportunity after processors neutralize potentially dangerous branch predictor states, allowing attackers to re-poison the branch predictor and leak sensitive data.

The researchers, Daniël Trujillo and Mengjia Yan, demonstrated that the TONTOU attacks work on both Intel and AMD processors, targeting mitigations designed to protect against Spectre-style attacks. By injecting unprivileged programs with high-frequency timer interrupts during the post-neutralization window, the attackers can divert control flow to execute an interrupt handler that re-poisons branch predictor structures like the return stack buffer (RSB) or branch history buffer (BHB).

This causes protected branches to speculatively jump to a disclosure gadget, leaking kernel data through side channels. The researchers showed that their attacks worked on Intel Cascade Lake Refresh and Arrow Lake processors and AMD Zen 2 and Zen 4 chips. They also managed to break Linux's kernel address space layout randomization (KASLR) and locate specific secrets like the etc/shadow file, which contains the root password hash.

While the attack is not simple and took about 18 minutes to execute, it is practical and can affect multi-tenant container platforms where ordinary user space programs share the kernel with victims. Intel and AMD have not committed to addressing the issue with kernel patches, though AMD stated it would develop a response.