Skip to content

Minutes 03 Apr 2025

Paul Albertella edited this page Apr 3, 2025 · 1 revision

Host: Paul Albertella

Participants: Leonardo Rosesetti, Florian Wuehr, Igor Stoppa, Daniel Krippner, Gabriele Paoloni, Pete Brink

Agenda

  • Updates on documents
  • Building on the ‘Minimal use case’
  • RAFIA, STPA and TSF

No real updates on documents

Minimal use case

Igor: More formalised form of what we’ve been trying to describe

  • So a buildable example of a Linux ‘hello world’ for analysis
  • Objective evidence of some of the things we have been saying
    • Intricate dependencies between subsystems
    • This applies even to the the simplest program
    • Local memory caches are used per-core

Benefits:

  • Clear test bed to illustrate the problem space
  • A way to counter claims that the problem can be simplified to ignore some of these ‘inconvenient truths’
  • A way to open the ‘black box’ and test our assumptions

Pete: Based on this, we could ask whether some of the complexity we are seeing is actually necessary - or could we create a configuration that omits (some of) it.

Igor: We don’t have to solve / remove it all, but we do need to be aware of it, or make it configurable where it is not

  • The kernel has mostly been developed with desktop or server performance in mind

Describing it comprehensively would give us an idea of what we need to deal with

Paul: Also we may want to make use of some of the performance features

Igor: First goal was to make it evident why some of those higher order solutions are not credible if the lower level aspects are not credible

  • Useful output would be a list of aspects that you need to take into account
  • Not claiming to be comprehensive, but a minimal set to be addressed
  • Not just a list, but a set of dependent aspects

Paul: An approach to describing these things is to use STPA

  • Adds the idea that not all problems make a system unsafe
  • Gab: Challenge we found when trying to apply STPA to kernel was that you needed a detailed specification of the functionality you are analysing
  • Igor: Perhaps we can take the outputs of the minimal use case analysis and apply STPA to it?

Next time:

  • Look at questions / responses for Pete’s list of review criteria

Clone this wiki locally