DRIFT: Debug-based Trace Inference for Firmware Testing
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Binary firmware fuzzing has garnered attention in recent years. Compared to source-code-based approaches, binary approaches require less semantic information and are therefore more applicable. This is particularly relevant in firmware analysis, as most firmware vendors distribute only binaries, withholding source code due to proprietary concerns.Pivoting away from the traditional hardware-in-the-loop (HiL) methodology, researchers are exploring more efficient ways to engage real hardware for fuzzing. However, existing approaches have inherent drawbacks, such as reliance on high-end hardware features, inability to recover complete coverage, and slow execution speeds. We propose DRIFT, a novel approach for on-device binary firmware testing that follows the semihosting methodology. DRIFT addresses all the aforementioned drawbacks. The core insight of DRIFT is to use the Debug Monitor (DM) for firmware fuzzing. DM is a Arm Cortex-M CPU feature that allows triggering interrupt when a breakpoint is hit. Through chaining the DM interrupts, DRIFT is able let firmware to trace itself. This self-tracing approach minimizes interference from the workstation, significantly boosting fuzzing performance.We designed DRIFT to be highly flexible, accommodating a number of hardware resource limitations. When applied to new firmware, DRIFT discovered three previously unknown bugs that were not identified by existing binary fuzzing techniques. Furthermore, DRIFT outperforms all state-of-the-art binary firmware fuzzers in terms of speed and fidelity, trailing only SHiFT, an approach that requires source code.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.014 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.002 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.002 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it