MEI: A Light Weight Memory Error Injection Tool for Validating Online Memory Testers

摘要

Lots of studies have shown that memory hardware error rates are orders of magnitude higher than previously reported. In order to fight with these memory hardware errors, many memory testing tools have been developed, especially software level online memory testers, which means these memory testers implemented in software can work with the OS (operating system) at the same time. However, validation of these online memory testers is a hard work. Since the real broken memory chips is hard to find and using a virtual machine to do this work is really complex. So we have developed a memory error injection tool - MEI (Memory Error Injection), which is implemented in software on Linux platform and easy to use. The core function of MEI is implemented in the form of a Linux kernel module. MEI also provides some user space tools for memory errors injection and manipulation. The testers to be validated need to use the read interfaces provided by MEI. We have used two exists memory testers, the modified RAMpage and COMeT+ model, to validate the effectiveness of MEI. The results of experiments have shown that these two testers could detect the emulated memory hardware errors injected by MEI effectively.