ContextQMD
Back to Tock

Tock Core Notes 2022-01-21

doc/wg/core/notes/core-notes-2022-01-21.md

latest3.5 KB
Original Source

Tock Core Notes 2022-01-21

Attendees

  • Amit Levy
  • Alexandru Radovici
  • Brad Campbell
  • Leon Schuermann
  • Jett Rink
  • Philip Levis
  • Johnathan Van Why
  • Hudson Ayers
  • Vadim Sukhomlinov
  • Pat Pannuto
  • Branden Ghena
  • Alyssa Haroldsen

Updates

  • JVW: libtock-rs. New script for running elf2tab and running binaries via QEMU and tockloader. Useful for testing with 2.0.
  • Alyssa: libtock-rs. Updates to build system for assembler and archive to work with specific addressing requirements.
    • Could we just use LLVM tools?
    • Might be missing assembler.
  • Ran into assembler bug with comments on move instructions.

ufmt experience

  • Hudson: considered vendoring ufmt into libtock-rs to replace core formatting libraries to reduce binary size and increase performance in some cases.
  • Ported ti50 applications to use ufmt on libtock-rs.
  • Experimented with using it in the kernel as well.
  • Findings
    • Doesn't support all of the same format specifiers. Supports only normal, debug, and :#.
    • No support for hex numbers.
    • No support for many number formatting and padding tools.
    • Difficult to write independent code that works with both core and ufmt. Would have to limit to just the very limited interface of ufmt.
    • For ti50 apps, saved ~13 kB!
  • Now working on ufmt-extended that adds some important features, see what code size effects are.
  • Is this viable for apps?
  • Looking to support most formatting for hex numbers.
  • Seems like it could work for most apps.
  • Issue: if library doesn't use it, might have to include both corefmt and ufmt.
  • ufmt is not "pay for what you use", the entire binary size is added even if you don't use certain features.
  • If you use a feature that doesn't exist, does it fail at compile time?
    • Yes.
  • Any attempt to do the formatting in the kernel?
    • Debug hard to do since the type is complex.
    • For simple types, would be feasible, but would result in more syscalls.
    • Could overlap with console syscall?
      • Still might require many more syscalls.
      • Might want to use formatting more generally.
      • Would need to pass the structure of the thing to be printed.
    • Macro implementation, how much could be done at macro invocation time?
      • Not clear yet.

AppID

  • Phil: state machine in place for loading multiple processes with credentials and without.
  • Dummy implementations for async checks (placeholders until RSA implementations exist).
  • Three changes to process:
    • New queue init task fn
    • Two credentials functions: mark as pass, mark as fail
  • What does toolchain look like for working with credentials?
    • Brad: Adding to tockloader would be straightforward.
    • elf2tab can add a header specifying the footer, or tockloader can insert that.
  • Hudson: if someone does not want to use this, what do they pay for in code size?
    • Goal is code is structured so all unused features are elided.

Process Slice

  • Jett: remove ProcessBuffer layer in exchange for a check on access that makes sure no aliasing.
  • Any invalid mutable access to a process slide would fail.
  • Checks would be at runtime.
  • Phil: we thought about this at allow time, not necessarily use time. Doing at use time could lead to unusual failures in certain cases. Checking at allow-time adds some overhead (100s of cycles).

Testing for Tock