BlockHammer: Preventing RowHammer at Low Cost by Blacklisting Rapidly-Accessed DRAM Rows π
This work introduced Blockhammer, a rowhammer mitigation mechanism of both scalability with worsening rowhammer vulnerability and compatibility with commodity DRAM chips.
Strengths
Besides mitigation effect, performance and energy are also very important and is well considered in this paper with rigorous experiments.
- BlockHammer introduces very low performance and DRAM energy overheads.
- BlockHammer significantly increases benign application performanceand reduces DRAM energy consumption.
This work also includes rigorous security proof and mathematically represent all possible access patterns.
There are also many evaluations over other approaches, even with category information. It is really good related work summarizing and comparison.
Weaknesses
- There is tradeoff between recall and precision in the blockhammer machenism, and I did not find adequate discusion or experiments over how to balance these two metric.
Can you do better?
Blockhammer is based on access pattern, which means it depends on the attacker. In other words, it is negative instead of positive when making defense. Even worse, the attacker may train some adversarial model to attack the DRAM, by causing high defense overhead.
I would find some more positive defense methods.
I would do more experiments or mathematics to calculate what is the better tradeoff strategy to defense rowhammer.
Takeways
Blockhammer is really good at grabbing the following two points:
- Scalability with worsening RowHammer vulnerability.
- Compatibility with commodity DRAM chips: no modifications to DRAM circuitry and no depedence on internal information of DRAM.