Breaking Lab srl
A Race Against the Clock Inside Your CPU
Modern processors are extraordinarily fast, and that speed comes with a price. Ever since the Spectre and Meltdown vulnerabilities shook the computing world in 2018, hardware vendors and operating system developers have worked together to layer on mitigations designed to prevent attackers from exploiting speculative execution — the technique CPUs use to stay busy by guessing which instructions are coming next. Years of patches later, many engineers assumed the worst was behind them. A new discovery from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) suggests that assumption was premature.
How INTERRUPT INJECTION Works
Researchers Daniël Trujillo and Mengjia Yan identified what they are calling INTERRUPT INJECTION, a technique that targets a fleeting but critical window inside a processor’s execution pipeline. Existing Spectre v2 defenses work by sanitizing, or flushing, the branch predictor — a CPU component that predicts which code paths will be taken — before the operating system kernel gets to use it. The idea is to ensure no attacker-controlled predictions linger to influence privileged kernel operations.
The problem is that sanitizing the branch predictor and the kernel actually using it are two distinct, sequential events. INTERRUPT INJECTION exploits the tiny gap between them. An unprivileged userspace program can deliberately time a hardware interrupt to fire precisely in that window. When the interrupt fires, it gives the attacker the opportunity to re-poison the branch predictor after the defense has already run but before the kernel has begun executing. In effect, the mitigation is undone at the last possible moment without the kernel ever knowing.
The researchers validated the attack on an AMD Zen 2 machine running Linux kernel 6.14, with every default Spectre v2 mitigation enabled. No elevated privileges were required — the attack was executed entirely from an ordinary unprivileged process, making it particularly dangerous in multi-user or cloud-hosted environments where multiple workloads share the same hardware.
Why This Matters Beyond the Lab
The practical implications are significant. Cloud computing platforms are among the most exposed environments: virtual machines belonging to different customers frequently run on the same physical hardware, with the kernel serving as the critical isolation boundary. An attacker who can run unprivileged code — for instance, through a compromised container or a rented virtual machine — could potentially use INTERRUPT INJECTION to leak sensitive data across that boundary, including cryptographic keys, passwords, or other memory belonging to co-located processes or the kernel itself.
The fact that the attack works against a fully patched, up-to-date Linux system is especially noteworthy. It demonstrates that the current generation of Spectre v2 defenses, while effective against the original attack vector, did not account for this particular timing interaction. Hardware manufacturers and Linux kernel maintainers will now need to investigate whether additional mitigations — potentially at both the software and microcode levels — are necessary.
The Broader Spectre of Hardware Security
Intel and AMD have not yet issued formal public responses to the INTERRUPT INJECTION findings, and no CVE number has been published for this specific technique at the time of writing. The disclosure follows established responsible-disclosure norms, meaning both chip vendors and the Linux kernel security team are expected to have been notified in advance of publication.
This discovery is a sobering reminder of how difficult hardware-level security is to get right. Unlike software bugs, vulnerabilities rooted in CPU microarchitecture cannot simply be patched with a single update; fixes often require microcode changes, kernel modifications, and in some cases performance trade-offs that affect millions of systems. The INTERRUPT INJECTION attack also highlights a broader pattern: as researchers systematically probe the defenses built in response to Spectre, they continue to find edge cases that the original mitigations did not fully address. As long as CPUs rely on speculative execution for performance — and there is no sign that will change — this cat-and-mouse game between attackers and defenders is likely to continue for years to come.
