Tock Meeting Notes 2024-09-06

Attendees

• Branden Ghena
• Brad Campbell
• Johnathan Van Why
• Amit Levy
• Ben Prevor
• Tyler Potyondy
• Pat Pannuto
• Hudson Ayers

Updates

• Tyler: Doing some research in to low power Tock, specifically using the type system.
  • Default kernel using 1.2 mA on nRF52. Doing nothing except running UART for process console. Otherwise it is about 200 uA.
  • Using a $90 power monitor (Power Profiler Kit 2).
• Amit: Hacking on SMA Q3 smartwatch. (https://shop.espruino.com/banglejs2)
  • Exposing some issues around buffer management in HILs and screen stack.
  • Screen is unique with its own special color format.
  • Question: where to bring up issues with screen?
    • Brad: Ideally, the tracking issue should be the hub. But any screen changes are fairly difficult, so helpful to know what the goal/intent is before starting a bunch of changes.
• Hudson: Still working on nightly asm const PR.
  • Still working on a build.rs/cargo test solution that is suitable.
  • Difficult to manage the precedence of how options are selected.
  • Brad: I still think a proper build crate for Tock would make a lot of subtlety much more clear.

Leasable Buffers

• SubSlice in upstream kernel, PacketBuffer in the works.
• Would be helpful to use SubSlice in SPI HIL.
• Question: what is the relationship between PacketBuffer and SubSlice?
  • API: more adjustable slice and unslice with PacketBuffer. Can take/hide arbitrary bytes with PacketBuffer.
  • PacketBuffer provides guarantees about space at compile time.
• Question: what is the underlying representation of PacketBuffer?
  • SubSlice is just a Rust slice and an active range.
  • PacketBuffer is similar to SubSlice, has slice and active range.
• How far are we from being able to try out PacketBuffer?
  • Unclear.
• How aggressively should we switch to Leasable Buffers?
  • Brad: Let's do it! https://github.com/tock/tock/issues/3504
  • Branden: We don't need to wait for PacketBuffer.
• Hopefully easier to switch SubSlice -> PacketBuffer than [u8] to SubSlice.
• It can be tricky to switch to SubSlice in a meaningful way. Easy to escape hatch.
• Fairly easy to switch SPI, but not getting full benefit from SubSlice.
• Two aspects: enabling new capsules to be able to use SubSlice, and having good reference code to copy from.
  • Hopefully switching the HIL is the key to allow new capsules to be able to embrace SubSlice.
• There are three versions: SubSlice, SubSliceMut, SubSliceMutImmut to support different mutability of buffers.
  • Callbacks make using SubSliceMutImmut hard, need to ensure you get a mutable slice back. Leads to runtime checks.

Updating Rust

• CHERI needs out-of-tree compiler. That version is slightly behind latest Rust.
• How often do we update Rust?
• At the point of removing nightly features.
• We could want stable naked functions.
• Our current cadence is we update roughly every two months plus when something new is added we want.
  • Compiler changes might be slowing down, but there are a lot of other tools we use that see more frequent changes.
• Could use nightlies that correspond to Rust stable versions.
  • https://forge.rust-lang.org/
  • Seems like a reasonable option.

Release???

• How close are we to using treadmill for testing for a release?
  • Basic functionality there (reading console output).
  • Issues around starting jobs.
  • Some basic tests are there.
• Seems like a good time based on the current PRs. More features than fixes or low-level changes.