Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
---
|
| 2 |
-
license:
|
| 3 |
task_categories:
|
| 4 |
- text-generation
|
| 5 |
language:
|
|
@@ -21,22 +21,81 @@ size_categories:
|
|
| 21 |
|
| 22 |
# Linux Kernel Assembly → Explanation Dataset
|
| 23 |
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
license: gpl-2.0
|
| 3 |
task_categories:
|
| 4 |
- text-generation
|
| 5 |
language:
|
|
|
|
| 21 |
|
| 22 |
# Linux Kernel Assembly → Explanation Dataset
|
| 23 |
|
| 24 |
+
> **Upload in progress.** Additional batches are still being added. The final dataset will contain records derived from 90K+ BFC/BIC commit pairs.
|
| 25 |
+
|
| 26 |
+
A dataset of disassembled Linux kernel functions paired with structured natural-language explanations, built from the [Linux Commits Dataset (Zenodo 10654193)](https://zenodo.org/records/10654193).
|
| 27 |
+
|
| 28 |
+
Each row corresponds to a single compiled function extracted from a `.c` or `.h` file touched by a **Bug-Fix Commit (BFC)** or **Bug-Introducing Commit (BIC)** in the Linux kernel git history. Source files are compiled with `gcc -O2 -g -fno-inline -fno-omit-frame-pointer` and disassembled with `objdump -d -S` (AT&T syntax, source-interleaved).
|
| 29 |
+
|
| 30 |
+
Intended for training or evaluating models on **assembly code understanding**, specifically, generating natural language explanations from disassembled Linux kernel functions.
|
| 31 |
+
|
| 32 |
+
---
|
| 33 |
+
|
| 34 |
+
## Schema
|
| 35 |
+
|
| 36 |
+
| Column | Type | Description |
|
| 37 |
+
|---|---|---|
|
| 38 |
+
| `bfc_hash` | `string` | Hash of the bug-fix commit in the pair |
|
| 39 |
+
| `bic_hash` | `string` | Hash of the bug-introducing commit in the pair |
|
| 40 |
+
| `commit_hash` | `string` | The specific commit this function was extracted from |
|
| 41 |
+
| `commit_type` | `string` | `"bfc"` (bug-fix commit) or `"bic"` (bug-introducing commit) |
|
| 42 |
+
| `filename` | `string` | Source file path within the kernel tree (e.g. `kernel/sched/core.c`) |
|
| 43 |
+
| `func_name` | `string` | Name of the disassembled function |
|
| 44 |
+
| `asm` | `string` | `objdump -d -S` output for one function (AT&T syntax, source lines interleaved) |
|
| 45 |
+
| `explanation` | `string` | Structured description: commit type, function name, author, date, commit message, and role |
|
| 46 |
+
| `commit_message` | `string` | First 512 characters of the commit message |
|
| 47 |
+
|
| 48 |
+
---
|
| 49 |
+
|
| 50 |
+
## Usage
|
| 51 |
+
|
| 52 |
+
```python
|
| 53 |
+
from datasets import load_dataset
|
| 54 |
+
|
| 55 |
+
ds = load_dataset("theelderemo/linux-asm-pairs", split="train")
|
| 56 |
+
print(ds["asm"])
|
| 57 |
+
print(ds["explanation"])
|
| 58 |
+
```
|
| 59 |
+
|
| 60 |
+
Filter to only bug-fix commits:
|
| 61 |
+
```python
|
| 62 |
+
bfc_only = ds.filter(lambda x: x["commit_type"] == "bfc")
|
| 63 |
+
```
|
| 64 |
+
|
| 65 |
+
---
|
| 66 |
+
|
| 67 |
+
## Data Pipeline
|
| 68 |
+
|
| 69 |
+
1. BFC/BIC pairs sourced from `bfc_bic.csv` in [Zenodo record 10654193](https://zenodo.org/records/10654193)
|
| 70 |
+
2. Full diffs fetched from a local bare clone of the Linux kernel git repo (snapshot: 2023-11-12)
|
| 71 |
+
3. Changed `.c` / `.h` files compiled with `gcc -O2 -g -fno-inline -fno-omit-frame-pointer`
|
| 72 |
+
4. Functions extracted via `objdump -d -S --no-show-raw-insn`
|
| 73 |
+
5. Explanations constructed from Perceval-parsed commit metadata
|
| 74 |
+
|
| 75 |
+
---
|
| 76 |
+
|
| 77 |
+
## License
|
| 78 |
+
|
| 79 |
+
Source assembly is derived from the Linux kernel, which is licensed under **GPL-2.0**. This dataset inherits that license. See [kernel.org/doc/html/latest/process/license-rules.html](https://www.kernel.org/doc/html/latest/process/license-rules.html) for details.
|
| 80 |
+
|
| 81 |
+
---
|
| 82 |
+
|
| 83 |
+
## Citation
|
| 84 |
+
|
| 85 |
+
If you use this dataset, please also cite the upstream source:
|
| 86 |
+
|
| 87 |
+
```bibtex
|
| 88 |
+
@dataset{linux_commits_2023,
|
| 89 |
+
title = {Linux Commits Dataset},
|
| 90 |
+
year = {2023},
|
| 91 |
+
url = {https://zenodo.org/records/10654193}
|
| 92 |
+
}
|
| 93 |
+
|
| 94 |
+
@dataset{theelderemo_linux_asm_pairs_2026,
|
| 95 |
+
author = {theelderemo},
|
| 96 |
+
title = {Linux Kernel Assembly → Explanation Dataset},
|
| 97 |
+
year = {2026},
|
| 98 |
+
url = {https://huggingface.co/datasets/theelderemo/linux-asm-pairs},
|
| 99 |
+
note = {Derived from the Linux Commits Dataset (Zenodo 10654193). Assembly extracted via GCC + objdump from BFC/BIC commit pairs in the Linux kernel git history.}
|
| 100 |
+
}
|
| 101 |
+
```
|