CAREER: Architectural Support for Automated Software Debugging

鉴于其日益增长的复杂性，现代软件系统饱受软件缺陷的困扰，这些缺陷通常被称为错误。在观察到程序失败后，软件开发人员通常需要花费大量的努力来定位缺陷。由于当时片上资源有限，对调试的传统体系结构支持仅限于断点和观察点等基本原始功能集。随着半导体技术的进步，资源限制不再是一个令人担忧的问题，实现更强大的架构支持成为可能，以简化软件调试。在这项研究中，开发了一种新颖的软硬件集成方法来自动定位软件缺陷，目的是开发一种计算机，能够自动定位顺序或并行程序中的教师代码，并可能生成对缺陷的修复。相比之下，本研究引入了新颖的架构支持：Bug检测报告潜在Bug，Bug隔离根据潜在Bug与程序故障之间的因果关系找到相关Bug，Bug验证生成对隔离Bug的快速修复，从而形成一个完整的自动化调试过程。顺序程序和并行程序中的错误都是本研究的目标。对于并行程序，研究了事务内存编程模型下的线程交互，并针对并发错误开发了新的自动调试方案。该研究还包括新的架构支撑的原型，以评估其与现实世界应用的有效性。

Given their ever increasing complexity, modern software systems are plagued with software defects, commonly known as bugs. It usually takes significant amount of efforts for software developers to locate the defects after a program failure is observed. Due to the limited on-chip resource at the time, traditional architectural support for debugging was limited to a basic set of primitive functions like breakpoints and watchpoints. With the advances in semiconductor technology, the resource constraint is less of a concern and much more powerful architectural support becomes possible to be implemented to ease software debugging. In this research, novel software-hardware integrated approaches are developed to automatically pinpoint software defects and the aim is to develop a computer that can automatically pinpoint the faculty code in either sequential or parallel programs and potentially generate a fix to the defect.Previous work on architectural support for debugging mainly focused on one aspect of debugging activities including faithfully reproducing program failures or detecting potential bugs. In comparison, this research introduces novel architectural support for: bug detection to report potential bugs, bug isolation to find the relevant bugs based on cause-effect relationship between the potential bugs and the program failure, and bug validation to generate quick fixes to the isolated bugs, thereby forming a complete process of automated debugging. Bugs in both sequential and parallel programs are the target in this research. For parallel programs, the research investigates thread interaction under the transactional memory programming model and develops novel automated debugging schemes for concurrency bugs. The research also includes the prototype of the novel architectural supports to evaluate their effectiveness with real-world applications.