基于质因数分解的控制流错误检测方法

软件控制流错误检测技术是防止由于单粒子翻转事件而造成程序错误运行的有效手段之一,其方法主要是将编译时计算的签名值同运行时生成的签名值进行比较.因此,体现基本块间依赖关系的签名值的表示方法决定软件控制流错误检测概率的高低.为了在低检测代码开销的情况下提高错误检测概率,结合图论理论提出一种汇编层的控制流检测方法.该方法运用整数可唯一分解为素数乘积的代数基本定理表示基本块间依赖关系的要素,以此作为生成签名值的依据,将其插入以基本块为单位的检测指令中.实验表明,此算法的性能评价指标优于CFCSS等具有代表性的软件控制流检测方法,具有很强的实用价值.

Detecting errors in program flow based on prime factorization

An error detection technique for software control flow,where a signature value set on compilation is compared with one derived at run time,is an effective way to prevent break-downs of program operations because of single event upsets.However,the probability of providing an effective error detection rate depends on where checks on the signature values are placed,which must correlate to basic blocks of code.In order to improve the error-detection rate at the lowest cost in detection codes,a control flow detection method based on an assembling layer is presented in combination with graph theory.The signature values are integers,and can be uniquely expressed as the products of prime numbers.These prime numbers can be chosen to reflect the dependencies of the basic blocks,and their relationships are then used to generate signature values.We employed this concept to determine the signature values to insert into detection instructions where each basic block is taken as a unit.The experimental study shows that the performance index of this proposed algorithm is better than that of CFCSS and other classic algorithms.