CEDARtools.SmartTrace for WCET analysis

Measured task runtimes on real silicon feed AbsInt TimeWeaver for hybrid WCET analysis - the measurement half of a two-sided argument.

In brief

Worst-case execution time on multi-core targets is hard to establish by static analysis alone, and hard to establish by measurement alone. The practical approach is hybrid: static analysis provides the structure and the bounds, measurement provides the real execution times of real paths on real silicon. SmartTrace delivers that measurement side: task runtime profiling from hardware trace.

Which trace source carries it depends on the target: Infineon^® AURIX^™ exposes execution through the MCDS mechanism (TC3xx and TC4xx), Arm^® cores through CoreSight^™ ETM, and RISC-V^® cores through processor trace (N-Trace and E-Trace). The mechanisms differ in what they report and at what rate, so what is documented per target - and with which status - is answered per configuration.

Continuous observation records every run of the task on the running target, so a rare maximum has a chance of being seen; the raw trace of the longest observed run of each task is what feeds the static analysis that produces the WCET estimate.

The hybrid workflow

The measured timing feeds AbsInt TimeWeaver for hybrid WCET analysis. That combination was presented jointly with AbsInt at ERTS² 2024 - the paper is on the publications page. The combined workflow can contribute measured execution and interference observations to a project argument addressing the objectives of EASA AMC 20-193; sufficiency remains a project determination.

Where the measured times land: AbsInt's a3 Advanced Analyzer with a WCET estimate for a task entry point and the per-function call graph below it. The screenshot shows an example project shipped with the tool.

What the measurement gives you

The longest runs, kept - not everything. The point of the measurement is not a full recording; it is the worst case that actually occurred. While the target runs, each task’s execution time is watched live, and every time a task exceeds its own previous maximum, the raw trace of that run replaces the earlier one - one buffer slice per task, several tasks and interrupt handlers in parallel. At the end of a campaign what remains is exactly what a WCET argument needs as input: the longest observed run of each task, as raw trace. How that works in the hybrid workflow
Real paths on real silicon. Passive observation of the running system - no measurement runtime in the application instruction stream.
Task runtime profiling, named precisely. The underlying mechanism differs per platform - MCDS on AURIX^™ gives a different view than CoreSight^™ on Arm^® - and we would rather name what each configuration does than stretch a word.
A stated retention record. The configuration determines whether raw trace, reconstructed events, aggregated metrics or only final reports are retained; the report states which artifacts remain available for reproduction or independent review.

Tool qualification

Timing evidence that supports a certification argument raises the same question as any other verification tool output: what qualification evidence exists for the tool that produced it - tool qualification data under DO-330, a tool confidence level argument under ISO 26262. If you are interested in a tool qualification kit for CEDARtools.SmartTrace WCET analysis, contact our support - we will discuss scope and certification context with you.

The product in full · Timing & WCET in detail · Target support, answered per configuration