# Description
This dataset consists of the memory access behavior of various applications, including a mix of both normal applications and Rowhammer attacks, and associated code. The memory access data is generated using Gem5 v20.1.0.5 [9]. We use Gem5 to simulate a 4-core system with Linux OS in full-system mode. The data can be used for Rowhammer research or any other memory related research purpose.


# Citations
If you use this dataset or any part of it, please cite one of our papers:

[1] B. K. Joardar, T. K. Bletsch and K. Chakrabarty, "Learning to Mitigate Rowhammer Attacks," 2022 Design, Automation & Test in Europe Conference & Exhibition (DATE), 2022, pp. 564-567, doi: 10.23919/DATE54114.2022.9774703.

[2] B. K. Joardar, T. K. Bletsch and K. Chakrabarty, "Machine Learning-based Rowhamemr Mitigation," in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2022 (under review currently).


# What is in this dataset
The dataset contains a mix of both Rowhammer attacks and normal applications. 

Rowhammer: 
Rowhammer is a DRAM vulnerability that occurs when the same set of row(s) is/are accessed multiple times. The memory access behavior of the Rowhammer attacks (referred as G-hammer and TRRespass in [1]) were generated by modifying the attack code from [3]. The adversarial and Blacksmith attacks are generated using the adversarial attack generator following the DRAM operating principles. 

Normal applications: 
The normal applications belong to Splash-2[4], Parsec[5], Pampar[6], Spec2006[7] and Spec2017[8] benchmark suites. The memory access patterns for these applications are obtained from Gem5 full-system simulations.


# Folder-by-folder description
The folder 'n-sided' includes the memory acess patterns for the G-hammer and TRRespass attack [3][10]. The file names indicate the type of attack. For instance '7sided' indicates that we are hammering 7 rows in a bank.

The folder 'blacksmith' includes the memory access pattern of the blacksmith attacks [11]. The file names indicate the type of attack. For instance 'intense2_3sided'  refers to the attack where we hammer 3 rows with an intensity of 2. For more details about the parameters, please refer [1].

The folder 'adversarial' includes potential adversarial attacks for evaluation. Here, we consider burst attacks and irregular attacks. '3sided' refers to the attack where 3 rows are hammered.

The folder 'benign' includes the memory access patterns for the normal applications (from the Splash-2, Parsec, Pampar, Spec2006, and Spec2017 [4-8])

All experiments were done on a simulated 4-core system with linux OS and full system simulations on Gem5 [9].

The folder 'code' has the relevant code to read and process the Gem5 outputs. Each file in this folder decribes its purpose in its first three lines.


# Format of the data
Each file is of the form:

format: address	 bank	 rank	 row	 cycle

'Address' refers to the actual DRAM address being accessed; 'Bank','Rank', 'Row', are DRAM parameters and are related to the 'Address'. For details about these terms, please refer [1][10]. 'Cycle' indicates the time when the 'Address' was accessed.



# References
[3] M. Seaborn and T. Dullien, "Exploiting the DRAM rowhammer bug to gain kernel privileges," in BlackHat, 2015
[4] S. C. Woo wt. al., “The SPLASH-2 programs: characterization and methodological considerations,” in ISCA 1995, 24-36.
[5] C. Bienia, et. al., "The PARSEC benchmark suite: Characterization and architectural implications," in PACT, 2008, pp. 72-81.
[6] A. M. Garcia et. al, “PAMPAR: A new parallel benchmark for performance and energy consumption evaluation,” in CCPE. 2019.
[7] Standard Performance Evaluation Corp., "SPEC CPU 2006", https://www.spec.org/cpu2006/
[8] Standard Performance Evaluation Corp., "SPEC CPU 2017", https://www.spec.org/cpu2017/
[9] J. L. Power et. al., “The gem5 Simulator: Version 20.0+,” in arXiv:2007.03 152, 2020. 
[10] P. Frigo et al., "TRRespass: Exploiting the Many Sides of Target Row Refresh," in S&P, 2020, pp. 747-762.
[11] P. Jattke et. al., "BLACKSMITH: Scalable Rowhammering in the Frequency Domain," in S&P, 2022