Tock Meeting Notes 2025-03-05

Attendees

• Branden Ghena
• Brad Campbell
• Pat Pannuto
• Johnathan Van Why
• Amit Levy
• Viswajith
• Alexandru Radovici
• Kat Fox
• Leon Schuermann
• Hudson Ayers
• Tyler Potyondy

Updates

• None

Dynamic Process Loading

• TRD: https://github.com/tock/tock/pull/4338
• Adding comments in PR
  • Should dynamic store capsule match trait name?
  • Where are resources allocated, e.g. from app_loader capsule?
• Need to maintain the linked list structure of processes.
  • Need to call: setup(), then write() multiple times, then either abort() or load()
  • Should there be a "finalize" step that writes the total_length field to make the app valid?
  • What are the guarantees for maintaining the linked list and how is that enforced?
  • The requirements for linked list maintenance leaks into the interfaces.
    • We don't know how to build better abstractions right now that hide the linked list details.
  • A "finalize" operation might make it simpler to reason to about the state of flash.
• The store() and load() are separate traits, yet seem very linked.
  • They are very much linked. Expectation is that there is one app in flight being loaded.
  • Perhaps clearer documentation about setup() and load() is needed.
• Why is setup() separate from set_storage_client()?
  • The set_client() is just like any other set_client() in Tock. Multiple apps would need a virtualizer layer to handle multiple app-level clients.
• Why are these kernel capsules? Would capablities be a better solution?
  • Maybe. But where to store a process is quite sensitive.
  • The intent of DynamicBinaryStore and DynamicProcessLoad is to be generic to different backing storage mechanisms.