Design And Implementation Of Fault Injection Tools Targeted At Itanium 2 Processors

Nowadays computer systems have been widely applied in the fields of a number of key industries, such as civil aviation reservation systems, bank settlement systems and securities trading systems, where reliable and stable operation of the computer system plays a very crucial role. Itanium Family Processors, mainly targeted for mission-critical and data-intensive computing market, are often those the most usually employed servers in these areas. Intel claims that eighty of the biggest one hundred global telecommunication carriers have the deployment of Itanium servers. Therefore, the reliability evaluation for Itanium servers is very meaningful.In this paper, the method which observes system state under injected faults is employed to conduct reliability evaluation of a target system. Firstly, we analysis current fault injection methods and tools, and then design and implement a simulated fault injection tool in operating system kernel level based on the Itanium processor features. The tool can inject faults to 50 registers in the target processor. It supports not only a number of fault models such as fixed 0, fixed 1 and 1 bit flip, but also an adapter to set up fault interval as to simulate permanent, transient or intermittent faults.The tool is able to simulate a number of processor faults, but it is not accessible to complex structures inside the processor, and hence could not test them independently. Besides, simulated fault is not very close to those in reality. In this case, the author has developed another tool which injects faults in firmware layer. Based on IA-64 architecture firmware, it is able to inject a series of faults to target processors, such as processing-core failure, machine check errors in Cache/TLB/Register File or other inside-processor structures, cache contents fault etc. Finally, a series of fault injection experiments have been conducted to confirm the effectiveness of both tools and that they can play an important role in the practice of reliability evaluation.