TLB-pilot: Mitigating TLB Contention Attack on GPUs with Microarchitecture-Aware Scheduling

Architecture of TLB-pilot

demo

A simple example for using TLB_pilot

runtime.cuh

The library of TLB_pilot

TLB_contention_attack.cu

A attack for GPU TLB

malfunctioned_attack.cu

A malfunctioned attack kernel for GPU TLB

How to use TLB-pilot

• Including runtime.cuh in your source codes.
• Using following APIs to deploy TLB-pilot

/* Returns the number of thread blocks for the "filling phase".*/
new_block_num(int orig_block_num) 

/* Sends a message to the long-running kernel. */
void kernel_splitting_send(void)

/* It blocks execution until it receive a message from a short-running kernel. */
void kernel_splitting_receive(void)

• You can see examples in demo to learn more details.