Revisiting ARM Debugging Features: Nailgun and its Defense

Zhenyu Ning, Chenxu Wang, Yinhua Chen, Fengwei Zhang, Jiannong CaoIEEE TDSC 2023 · DOI

Introduction: The Shift in Debugging Paradigms

For decades, hardware debugging interfaces like JTAG were considered inherently secure because they required physical access to the device. However, as systems-on-chip (SoC) grew in complexity, ARM introduced the inter-processor debugging model starting with the ARMv7 architecture. This evolution allows one on-chip processor to debug another via memory-mapped interfaces, effectively removing the requirement for physical probes.

Comparison of traditional off-chip debugging versus the modern inter-processor debugging model.

While this model enhances flexibility for developers, it introduces a critical security oversight. The authentication signals governing these features (such as DBGEN and SPIDEN) were designed for the legacy physical-access model. They verify if the target processor is permitted to be debugged but fail to validate the privilege level of the host processor initiating the debug session. This "privilege blindness" allows a low-privilege kernel driver to seize control of high-privilege components like the Secure Monitor (EL3) or Hypervisor (EL2).

The Nailgun Attack Methodology

The researchers introduced Nailgun, an attack that exploits this architectural vulnerability without relying on software bugs. Nailgun utilizes two primary debugging components:

  1. Non-invasive Debugging (ETM): By leveraging the Embedded Trace Macrocell, attackers can observe the instruction flow of secure processes. This enables side-channel attacks on cryptographic implementations, such as AES, by inferring keys through observed memory access patterns.
  2. Invasive Debugging (ECT & EDITR): The attack uses the Embedded Cross Trigger (ECT) to halt a target processor. Once the target is suspended, the Debug Instruction Transfer Register (EDITR) is used to force the execution of specific instructions. Specifically, the dcps (Debug Change Privilege State) instruction is used to escalate the target's privilege to EL3, allowing for arbitrary code execution at the highest hardware privilege level.

Virtualization-Based Defense

To mitigate this threat, the authors proposed a lightweight virtualization-based defense. The core of this solution lies in ARM’s Stage 2 address translation, a feature typically reserved for hypervisors to isolate guest operating systems.

ARM memory translation stages, highlighting Stage 2 translation controlled by EL2.

By implementing a "gatekeeper" layer at EL2, the defense marks memory pages containing sensitive debug registers as invalid for non-secure operating systems. When a low-privilege OS attempts to access these registers, the hardware triggers a translation fault, which the hypervisor intercepts and blocks. This approach effectively secures the hardware-translation level without requiring changes to the underlying SoC hardware.

Evaluation and Results

The study evaluated Nailgun across 11 real-world platforms, including IoT devices (Raspberry Pi 3), cloud servers (ARM-based instances from Packet and Scaleway), and various mobile devices from Huawei, Samsung, and Xiaomi.

Key findings include:

  • Data Extraction: The researchers successfully extracted 128-bit AES keys in a single execution and retrieved raw fingerprint images from the TrustZone of commercial smartphones.
  • Full System Compromise: Arbitrary code execution was achieved at EL3 on development boards, granting total hardware control.
  • Performance: The virtualization defense proved highly efficient, with Linux benchmarks (Unixbench, Sysbench) showing a performance overhead of less than 1.1%.

Conclusion

Nailgun demonstrates that architectural features intended for maintenance can become potent attack vectors when isolation boundaries are not strictly enforced across all interfaces. Because many SoC manufacturers leave debug signals enabled by default for post-market analysis, millions of devices remain vulnerable. While the proposed virtualization defense is effective for ARMv8-A devices, legacy systems and future architectures like ARMv9 require continued scrutiny to ensure that debugging convenience does not come at the cost of fundamental security.

Summary updated

BibTeX
@article{ning2023revisiting,
  author = {Zhenyu Ning and Chenxu Wang and Yinhua Chen and Fengwei Zhang and Jiannong Cao},
  title = {{Revisiting ARM Debugging Features: Nailgun and its Defense}},
  journal = {IEEE Trans. Dependable Secur. Comput.},
  year = {2023},
  doi = {10.1109/tdsc.2021.3139840},
}
Figure