一种软硬件结合的容错技术研究

Research of an Error-tolerant Technique Based on Hybrid of Hardware and Software

单粒子翻转会使系统发生瞬时故障,基于控制流的错误检测技术,在运行时以基本块为单位由硬件或软件进行签名检验,判断运行是否出错.该技术嵌入签名信息使代码体积膨胀,系统性能下降,存在较多检测盲点.同时当前研究对错误检测后的系统恢复技术尚不成熟.提出一种软硬件结合的控制流容错技术.该技术覆盖大部分的检测盲点,并引入颗粒度校验模式优化代码密度,提高系统性能,同时采用硬件辅助的程序恢复技术来简化恢复流程.本文使用国产嵌入式CK-CPU作为实验平台,实验表明,这种软硬结合的容错技术能以较小的系统开销提高错误检测覆盖率,并针对完全校验模式实现错误恢复的功能.

Single event upsetswill cause transient faults of system, the error detection technique based on control flow judges whether errors exit through signature verify by hardware or software in basic blocks at runtime. Signatures will cause expansion of size of code and reduction of performance, and there are also some blind spots of detections. The technique of error-correction after error-detection is not mature. This paper proposes an error-tolerant technique bases on hybrid of hardware and software to cover most of the detection blind spots. To decrease the size of code and improve performance,the paper introduces granularity verify mode. Aim to correcterrors, the paper propose an error-correction technique based on hardware. We use CK-CPU as the platform experiment, experiment shows that the hybrid method will increase the coverage of error detection at lower system cost, and can correct the errors for full verify mode.