Mahsa Varshosaz will present her work on learning the shape of critical sections in the Linux kernel. Details below.
Mahsa Varshosaz, Postdoc.
Learning the Shape of Critical Sections in the Linux Kernel
In automatic repair of resource manipulation bugs, having information about which resources are protected can improve the quality of the suggested patches. In this work, our goal is to use learning to extract such information in the Linux kernel. As a first step, we investigate whether we can predict the shape of a critical section by learning from other existing critical sections. We use EBA to generate traces for files in the kernel as the main part of our dataset. We have designed a set of experiments on this dataset and the results include weights per line of code which suggest whether a line should be protected or not.