Detecting Unread Memory Using Dynamic Binary Translation

Abstract

Reading from uninitialized memory—that is, reading from memory before it has been written to—is a well-known memory usage error, and many static and dynamic tools verify that programs always write to memory before reading it. This work investigates the converse behaviour—writes that never get read, which we call “unread writes”. Such writes are redundant—at best, they do not perform any useful work; furthermore, work done to compute the values to be written could corrupt the program state or cause a crash. We present a novel dynamic analysis, implemented on top of the Pin dynamic binary translation framework, which detects instances of unread writes at runtime. We have implemented our analysis and present experimental data about the prevalence of unread writes in a set of benchmark applications.