SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2401.00563 · 2024

KernelGPT: Enhanced Kernel Fuzzing via Large Language Models

arXiv · PDF · Open in the Atlas

Code that ran

9 of 17 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 7 failures of our instrument; 8 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2401.00563.json · badge: /api/ran/2401.00563.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
ise-uiuc/kernelgpt canonical (classifier label)
classifier basis: census
9 ran · 8 no recorded run
FunctionStatusWhere it lives
extract_code_block Ran
ran; our contract draft was wrong, not the code
ise-uiuc/kernelgpt/spec-gen/llm_utils.py
code served (permissive licence) · get_code("cafe4e687de900be")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

# Analyze the function: extract_code_block takes a single parameter 'text' of type str
# The function splits text by newlines, looks for code block delimiters (``` or ```json),
# and extracts the content between them.

# Create a sample text with code blocks
text_with_code = """Here is some documentation.


print("Hello, World!")
x = 42
y = x + 1


More documentation text.

json
{
  "key": "value",
  "number": 123
}


Final notes."""

ARGS = {
    "text": text_with_code
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
extract_ioctl_function_name Ran
ran; our contract draft was wrong, not the code
ise-uiuc/kernelgpt/spec-gen/analyzer/process_output.py
code served (permissive licence) · get_code("0d9c3fed9939c48f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re

def extract_ioctl_function_name(code: str) -> str:
    """Extract the ioctl function name from the code."""
    pattern = '\\.(unlocked_)?ioctl\\s*=\\s*(\\w+)[,\\n]'
    match = re.search(pattern, code)
    if match:
        return match.group(2)
    return ''

# Setup code
code_sample_1 = ".ioctl = my_ioctl_handler,\n"
code_sample_2 = ".unlocked_ioctl = device_ioctl_func\n"
code_sample_3 = "some other code .ioctl = test_handler,"

ARGS = {
    "code": code_sample_1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
find_name Ran
ran; our contract draft was wrong, not the code
ise-uiuc/kernelgpt/spec-gen/find_utils.py
code served (permissive licence) · get_code("90a14820ad038a07")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
from pathlib import Path

# Construct valid arguments for find_name function
name = "model_weights"

path = "/home/user/projects/ml_models"

data_dict = {
    "model_weights": {
        "/home/user/projects/ml_models/checkpoint_v1.pt": b"model_data_1",
        "/home/user/projects/ml_models/checkpoint_v2.pt": b"model_data_2",
        "/home/user/data/ml_models/checkpoint_old.pt": b"model_data_3",
    },
    "config": {
        "/home/user/projects/config.json": b"config_data",
    }
}

ARGS = {
    "name": name,
    "path": path,
    "data_dict": data_dict,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
loguru — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_ops_type Ran
ran; our contract draft was wrong, not the code
ise-uiuc/kernelgpt/spec-gen/gen_utils.py
code served (permissive licence) · get_code("b361d3edf79bea45")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned OpsType. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

from enum import Enum, auto

class OpsType(Enum):
    SOCKET = auto()
    FS = auto()
    VIRT = auto()
    DRIVER = auto()

def get_ops_type(ops_path: str, ops_code: str=None):
    """Get the ops type from the ops name and path."""
    ops_path = str(ops_path)
    if 'linux/net' in ops_path:
        if not ops_code:
            return OpsType.SOCKET
        first_code_line = ops_code.splitlines()[0]
        if 'proto' in first_code_line:
            return OpsType.SOCKET
        else:
            return OpsType.DRIVER
    if 'linux/fs' in ops_path:
        return OpsType.FS
    if 'linux/virt' in ops_path:
        return OpsType.VIRT
    return OpsType.DRIVER

ARGS = {
    "ops_path": "linux/net/socket.c",
    "ops_code": "proto_handler()\nint main() {}"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type OpsType — a OpsType output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a OpsType output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_path_name Ran
ran; our contract draft was wrong, not the code
ise-uiuc/kernelgpt/spec-gen/gen_utils.py
code served (permissive licence) · get_code("537535ef80a7376f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import pathlib

torch.manual_seed(0)

# Analyzing the function:
# - path parameter: needs to be convertible to string via str()
#   The function checks if 'linux' is in the string representation
#   and either splits by 'linux/' or splits by ' '
# - replace_slash parameter: boolean that controls slash replacement

# Create a path object that contains 'linux' in its string representation
path_with_linux = pathlib.PurePosixPath('/home/linux/user/documents/file.txt')

# Create alternative path without 'linux'
path_without_linux = pathlib.PurePosixPath('/home/user/documents/file.txt')

ARGS = {
    'path': path_with_linux,
    'replace_slash': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_unique_name Ran
ran; our contract draft was wrong, not the code
ise-uiuc/kernelgpt/spec-gen/gen_utils.py
code served (permissive licence) · get_code("1765f746170fcd88")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

def get_path_name(path, replace_slash=True):
    if 'linux' in str(path):
        path_str = ''.join(str(path).split('linux/')[1:])
    else:
        path_str = str(path).split(' ')[0]
    if replace_slash:
        path_str = path_str.replace('/', '_')
    return path_str

def get_unique_name(name, path):
    path_str = get_path_name(path)
    return name + '#' + path_str

# Setup: analyze what get_unique_name uses
# - name: should be a string (used in concatenation with +)
# - path: passed to get_path_name which converts to str() and checks for 'linux' substring

# Create test arguments
test_name = "experiment"
test_path = "/home/linux/models/checkpoint.pt"

ARGS = {
    'name': test_name,
    'path': test_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
is_primitive_type Ran
ran, violated the contract we drafted
ise-uiuc/kernelgpt/spec-gen/output_spec.py
code served (permissive licence) · get_code("24ef7eee66b71029")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

ARGS = {
    "type_name": "int32"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_data Ran
ran; our contract draft was wrong, not the code
ise-uiuc/kernelgpt/spec-gen/find_utils.py
code served (permissive licence) · get_code("781825a17023e586")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import json
import tempfile
import os

# Create temporary JSON files for testing
temp_dir = tempfile.mkdtemp()

# Create first JSON file
file1_path = os.path.join(temp_dir, "data1.json")
with open(file1_path, 'w') as f:
    json.dump({"key1": "value1", "key2": 42}, f)

# Create second JSON file
file2_path = os.path.join(temp_dir, "data2.json")
with open(file2_path, 'w') as f:
    json.dump({"key3": "value3", "key4": [1, 2, 3]}, f)

# Define ARGS with list of paths to JSON files
ARGS = {
    "path_list": [file1_path, file2_path]
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
path_similarity Ran
ran, honoured the contract we drafted
ise-uiuc/kernelgpt/spec-gen/analyzer/process_output.py
code served (permissive licence) · get_code("d02df34683e889ea")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os

# path_similarity takes two string paths as arguments
# It splits them by os.sep, finds common and union of components
# Returns the ratio of common to total unique components

path1 = "/home/user/documents/file.txt"
path2 = "/home/user/downloads/image.png"

ARGS = {
    "path1": path1,
    "path2": path2,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
create_shared_linux_with_config Not yet run
unverified (harvested, no recorded run)
ise-uiuc/kernelgpt/spec-eval/run-specs.py
code served (permissive licence) · get_code("5b30bbd5da2cbcae")
How it ran

no invocation recorded.

debug_output Not yet run
unverified (harvested, no recorded run)
ise-uiuc/kernelgpt/spec-gen/analyzer/process_output.py
code served (permissive licence) · get_code("2636b92babc3ddd6")
How it ran

no invocation recorded.

find_type Not yet run
unverified (harvested, no recorded run)
ise-uiuc/kernelgpt/spec-gen/find_utils.py
code served (permissive licence) · get_code("b1230f5e4e37c060")
How it ran

no invocation recorded.

get_open_str Not yet run
unverified (harvested, no recorded run)
ise-uiuc/kernelgpt/spec-gen/output_spec.py
code served (permissive licence) · get_code("bf9453de6b993795")
How it ran

no invocation recorded.

get_shared_linux_with_config Not yet run
unverified (harvested, no recorded run)
ise-uiuc/kernelgpt/spec-eval/run-specs.py
code served (permissive licence) · get_code("34ce1ce60b0c72c6")
How it ran

no invocation recorded.

prepare_linux Not yet run
unverified (harvested, no recorded run)
ise-uiuc/kernelgpt/spec-eval/run-specs.py
code served (permissive licence) · get_code("c3570dbd1142e799")
How it ran

no invocation recorded.

query_ollm Not yet run
unverified (harvested, no recorded run)
ise-uiuc/kernelgpt/spec-gen/llm_utils.py
code served (permissive licence) · get_code("5506ea8f95924923")
How it ran

no invocation recorded.

run_eval Not yet run
unverified (harvested, no recorded run)
ise-uiuc/kernelgpt/spec-gen/eval_spec.py
code served (permissive licence) · get_code("4b04b35a66f40278")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 17 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Bugs in operating system kernels can affect billions of devices and users all over the world. As a result, a large body of research has been focused on kernel fuzzing, i.e., automatically generating syscall (system call) sequences to detect potential kernel bugs or vulnerabilities. Kernel fuzzing aims to generate valid syscall sequences guided by syscall specifications that define both the syntax and semantics of syscalls. While there has been existing work trying to automate syscall specification generation, this remains largely manual work, and a large number of important syscalls are still uncovered. In this paper, we propose KernelGPT, the first approach to automatically synthesizing syscall specifications via Large Language Models (LLMs) for enhanced kernel fuzzing. Our key insight is that LLMs have seen massive kernel code, documentation, and use cases during pre-training, and thus can automatically distill the necessary information for making valid syscalls. More specifically, KernelGPT leverages an iterative approach to automatically infer the specifications, and further debug and repair them based on the validation feedback. Our results demonstrate that KernelGPT can generate more new and valid specifications and achieve higher coverage than state-of-the-art techniques. So far, by using newly generated specifications, KernelGPT has already detected 24 new unique bugs in Linux kernel, with 12 fixed and 11 assigned with CVE numbers. Moreover, a number of specifications generated by KernelGPT have already been merged into the kernel fuzzer Syzkaller, following the request from its development team.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2401.00563")
get_code_for_paper("2401.00563")
have("2401.00563")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2401.00563.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2401.00563.svg)](https://syntology.ai/paper/2401.00563)

Connect an agent — have() is free.