Attack Can Bypass Spectre v2 Defenses on Intel and AMD CPUs

Overview

Researchers at MIT CSAIL have unveiled a new technique called Interrupt Injection, which can bypass existing Spectre v2 mitigations on both Intel and AMD CPUs. The attack exploits a timing gap between when the processor sanitizes its branch predictor and when the kernel uses it, allowing an unprivileged Linux program to re‑poison the predictor and leak sensitive kernel memory.

How Interrupt Injection Works

  • TONTOU class: Named after “Time‑of‑Neutralization to Time‑of‑Use,” similar to TOCTOU races in software.
  • Attack vector: By scheduling hardware interrupts with nanosecond precision, attackers can insert malicious instructions between mitigation and kernel use.
  • Demonstration: On AMD Zen 2 running Linux 6.14, researchers leaked kernel memory at 5.47 bytes/sec with 91.97% accuracy, enough to read /etc/shadow password hashes.
  • Persistence: Interrupts can fire almost anywhere, making them a powerful primitive for speculative execution attacks.

Research Findings

  • Affected CPUs: AMD Zen 1–4 processors (per AMD bulletin AMD‑SB‑7061). Intel CPUs also showed mispredictions, though no end‑to‑end leak was demonstrated.
  • Linux kernel patch: Commit “x86/bugs: Make Safe‑RET robust against interrupt injection” (June 2, 2026) fixes register state handling to prevent leakage.
  • Success rates:
    • Zen 2: 0.75% misprediction rate.
    • Intel Arrow Lake: 0.22%.
    • Cascade Lake Refresh: 0.037%.
    • Zen 4: No mispredictions observed.
  • Exploit chain: Attackers combined Interrupt Injection with Inception (CVE‑2023‑20569) to manipulate the return stack buffer.

Industry Responses

  • AMD: Confirmed vulnerability, issued bulletin, and shipped kernel patch.
  • Intel: Assessed behavior as covered by existing Spectre v2/BHI guidance, no new mitigations planned.
  • Arm: Acknowledged potential exposure but stated existing Spectre‑v2 and Spectre‑BHB mitigations already address risks.

Defensive Guidance

Administrators should:

  • Update Linux kernels: Ensure systems include the June 2 Safe‑RET patch.
  • Check vulnerability status: Review /sys/devices/system/cpu/vulnerabilities/spec_rstack_overflow.
  • Restrict unprivileged code execution: Especially on shared servers running affected CPUs.
  • Monitor speculative execution: Watch for anomalies in branch predictor behavior.

Expert in the Cloud Insight

Interrupt Injection demonstrates that Spectre defenses are not static—they can be undermined by creative timing attacks. The lesson for defenders is clear: mitigations must account for interrupt paths as part of speculative execution defenses. With researchers showing practical leaks on AMD Zen 2 and mispredictions on Intel, organizations should treat this as a high‑priority kernel update to safeguard sensitive environments.

Be the first to comment

Leave a Reply

Your email address will not be published.


*


This site uses Akismet to reduce spam. Learn how your comment data is processed.