The paradigm shift toward "New Space" has made necessary the adoption of Components- off-the-shelf (COTS) for space applications, including SoCs. While these systems offer the necessary computational power to execute modern algorithms and neural networks, they lack the radiation tolerance of legacy radiation-hardened hardware, shifting the respon- sibility of reliability from the hardware to the software layer. In safety-critical sectors such as the automotive and aerospace environments, faults induced by ionizing particles, specifically Single Event Upsets, present a significant threat to system integrity since they are capable of causing critical failures. The current state of the art struggles to address this challenge effectively. Physical radiation testing provides realistic data but is costly and can be destructive. On the other hand Simulation-based approaches such as RTL simulation offer great accuracy and reproducibility, but lack the throughput to execute complex software stacks. The Existing Software implemented Fault Injection (SWIFI) tools attempt to bridge this gap but often face a fundamental tradeoff: they either provide cycle-accurate simulation at an extremely low throughput, or employ inva- sive techniques that introduce high intrusiveness on a system, hindering the execution of large scale injection campaigns. This thesis proposes a Cross-Layer Kernel-Level Fault Emulation Framework for processor based systems designed to bridge this gap. By operating as a Loadable Kernel Module (LKM) on Linux it achieves near-native speeds, demonstrating a performance overhead of less than 12% across diverse workloads, while maintaining high controllability over the injection process. The framework was validated against traditional FPGA-based hard- ware injectors, showing consistent fault outcome distributions within a 4% margin. Fur- thermore, scalability tests confirm a linear speed-up when parallelized across multi-core architectures, enabling 100k injection campaigns to be completed in only a few hours. This provides developers a high-performance open-source solution to analyze software ra- diation resilience on processor based systems.
l cambio di paradigma verso il "New Space" ha reso necessaria l’adozione di compo- nenti COTS (Components-off-the-shelf) per applicazioni spaziali, inclusi i SoC. Sebbene questi sistemi offrano la potenza di calcolo necessaria per eseguire moderni algoritmi e reti neurali, mancano della tolleranza alle radiazioni tipica dell’hardware legacy "radiation- hardened", spostando la responsabilità dell’affidabilità dal livello hardware a quello soft- ware. In settori critici per la sicurezza come quelli automobilistico e aerospaziale, i guasti indotti da particelle ionizzanti, nello specifico i Single Event Upset (SEU), rappresentano una minaccia significativa per l’integrità dei sistemi poiché in grado di causare malfun- zionamenti critici. L’attuale stato dell’arte fatica ad affrontare con efficacia questa sfida. I test fisici di irraggiamento forniscono dati realistici, ma sono costosi e possono essere distruttivi, mentre gli approcci basati sulla simulazione, come la simulazione RTL, offrono grande accuratezza e riproducibilità, ma mancano del throughput necessario per eseguire programmi complessi. Gli strumenti esistenti di Software Implemented Fault Injection (SWIFI) tentano di colmare questo divario ma spesso affrontano un compromesso fonda- mentale: o forniscono una simulazione cycle-accurate con un throughput basso, oppure impiegano tecniche invasive che introducono un’elevata intrusività nel sistema, rendendo difficoltosa l’esecuzione di campagne di iniezione su larga scala. Questa tesi propone un framework di emulazione dei guasti a livello kernel e cross- layer per sistemi basati su processore, progettato per riempire questo vuoto implemen- tativo. Operando come Loadable Kernel Module (LKM) su Linux, il framework perme- tte di eseguire processi con velocità prossime a quelle native, dimostrando un overhead prestazionale inferiore al 12% su diversi carichi di lavoro, pur mantenendo alta la con- trollabilità sul processo di iniezione. Il framework è stato validato rispetto ai tradizionali iniettori hardware basati su FPGA, ottenendo dei risultati simili che rientrano entro un margine del 4% nel peggiore dei casi. Inoltre, i test di scalabilità confermano un aumento delle prestazioni lineare quando parallelizzato su architetture multi-core, consentendo di completare campagne da 100.000 iniezioni in poche ore. Ciò fornisce agli sviluppatori un framework open-source ad alte prestazioni per analizzare la resilienza alle radiazioni di software su sistemi basati su processore.
A cross-layer Kernel-level fault emulation framework for processor-based systems
CARBONETTI, ANDREA
2024/2025
Abstract
The paradigm shift toward "New Space" has made necessary the adoption of Components- off-the-shelf (COTS) for space applications, including SoCs. While these systems offer the necessary computational power to execute modern algorithms and neural networks, they lack the radiation tolerance of legacy radiation-hardened hardware, shifting the respon- sibility of reliability from the hardware to the software layer. In safety-critical sectors such as the automotive and aerospace environments, faults induced by ionizing particles, specifically Single Event Upsets, present a significant threat to system integrity since they are capable of causing critical failures. The current state of the art struggles to address this challenge effectively. Physical radiation testing provides realistic data but is costly and can be destructive. On the other hand Simulation-based approaches such as RTL simulation offer great accuracy and reproducibility, but lack the throughput to execute complex software stacks. The Existing Software implemented Fault Injection (SWIFI) tools attempt to bridge this gap but often face a fundamental tradeoff: they either provide cycle-accurate simulation at an extremely low throughput, or employ inva- sive techniques that introduce high intrusiveness on a system, hindering the execution of large scale injection campaigns. This thesis proposes a Cross-Layer Kernel-Level Fault Emulation Framework for processor based systems designed to bridge this gap. By operating as a Loadable Kernel Module (LKM) on Linux it achieves near-native speeds, demonstrating a performance overhead of less than 12% across diverse workloads, while maintaining high controllability over the injection process. The framework was validated against traditional FPGA-based hard- ware injectors, showing consistent fault outcome distributions within a 4% margin. Fur- thermore, scalability tests confirm a linear speed-up when parallelized across multi-core architectures, enabling 100k injection campaigns to be completed in only a few hours. This provides developers a high-performance open-source solution to analyze software ra- diation resilience on processor based systems.| File | Dimensione | Formato | |
|---|---|---|---|
|
Fault_Injection_Tool_Master_Thesis (1).pdf
accessibile in internet per tutti
Dimensione
1.88 MB
Formato
Adobe PDF
|
1.88 MB | Adobe PDF | Visualizza/Apri |
|
Executive_Summary_Master_Thesis_FIJ.pdf
accessibile in internet per tutti
Dimensione
551.37 kB
Formato
Adobe PDF
|
551.37 kB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/250457