Full Report
Timer interrupts reopen branch predictor poisoning window, with a working Zen 2 exploit to prove it
Analysis Summary
# Vulnerability: TONTOU (Time-of-Neutralization to Time-of-Use) Speculative Execution Attack
## CVE Details
- **CVE ID**: Not explicitly assigned in the article (Common for new academic research presented at DEF CON; often falls under broader Spectre v2 umbrellas like CVE-2017-5715 or CVE-2022-29900).
- **CVSS Score**: N/A (Research-stage vulnerability)
- **CWE**: CWE-1037 (Processor Optimization Removal or Modification of Security-critical Code) / CWE-142 (Properly Constrain Speculative Execution)
## Affected Systems
- **Products**: Intel and AMD CPUs; Linux OS.
- **Versions**:
- **Intel**: Cascade Lake Refresh and Arrow Lake architectures.
- **AMD**: Zen 2 and Zen 4 architectures.
- **Kernel**: Stock Linux kernels with default Spectre mitigations.
- **Configurations**: Multi-tenant environments (e.g., containers, shared cloud hosting) where an unprivileged attacker shares a kernel with a victim and has access to high-frequency timers.
## Vulnerability Description
TONTOU is a class of speculative execution attack that exploits a race condition between the **neutralization** of branch predictor state and the **use** of a branch (the "post-neutralization window").
Modern CPUs use mitigations like eIBRS (Intel) or Safe RET (AMD) to clear or isolate branch predictors when switching to privileged code (Kernel). TONTOU uses "interrupt injection"—scheduling high-frequency timer interrupts—to land exactly within this tiny window. Once the interrupt triggers after the security cleanup but before the victim branch executes, the interrupt handler is used to re-poison the Branch History Buffer (BHB) or Return Stack Buffer (RSB). This forces the CPU to speculatively execute a "disclosure gadget" that leaks sensitive data via side channels.
## Exploitation
- **Status**: PoC available. A working end-to-end exploit was demonstrated on **AMD Zen 2**.
- **Complexity**: High (Requires precise timing and the presence of specific disclosure gadgets in the kernel).
- **Attack Vector**: Local (Unprivileged user space to Kernel).
## Impact
- **Confidentiality**: High (Demonstrated bypass of KASLR and successful leakage of `/etc/shadow` root password hashes).
- **Integrity**: None.
- **Availability**: None.
## Remediation
### Patches
- **AMD**: Has committed to addressing the issue through upcoming Linux kernel patches.
- **Intel**: Stated they will not issue specific mitigations at this time, citing the high complexity and software-dependent nature of the attack.
- **Arm**: Indicated they do not actively protect against "passive leakage" via interrupt injections.
### Workarounds
- Disabling high-resolution timers for unprivileged users (may impact system performance/functionality).
- Implementing more frequent or "just-in-time" predictor sanitization, though this carries a significant performance penalty.
## Detection
- **Indicators of Compromise**: Difficult to detect via standard logs as it exploits hardware behavior.
- **Detection Methods**:
- Monitoring for unusual high-frequency timer interrupt activity from unprivileged processes.
- Performance counter monitoring for excessive branch mispredictions or cache misses associated with kernel entries.
## References
- MIT CSAIL Paper: hxxps[://]people[.]csail[.]mit[.]edu/mengjia/data/2026[.]USENIX[.]TONTOU[.]pdf
- Researcher Demo: hxxps[://]www[.]youtube[.]com/watch?v=L6yiLaS1EOc
- Original Reporting: hxxps[://]www[.]theregister[.]com/2026/08/07/mit_tontou_spectre_bypass/