BPF working group session @ IETF117

BPF Document Logistics

https://datatracker.ietf.org/doc/slides-117-bpf-bpf-document-logistics/

Current Process:

• Cross-posting to bpf at vger and IETF list.

• github.com/ietf-wg-bpf

  • Mirror of core documents in Linux tree
  • Contains boilerplate files
  • Auto-generated docs go to generated branch

• All core changes land in Linux tree under
  Documentation/bpf/standardization.

• There is an issue tracker on GitHub

Editors / authors / contributors / acknowledgments

• Experience from QUIC WG (Lars Eggert):
  • Used issues to figure out whether a text change was needed
  • Always have a link between an open issue and a PR

Filing submissions and issues:

• Concern regarding using non-free platforms (ie GitHub) for
  standardization (Christoph Hellwig)

  • Lars Eggert: QUIC WG set up quic-issues mailing list with
    notifications from GitHub for subscription without GitHub login
    and archiving in case GitHub goes away
  • Dave Thaler: No attachment to GitHub. Open to issues platform
    counter-proposal.
  • IETF errata approach not viable today.

• Presence on GitHub may set contributor expectations to accept PRs
  via GitHub (Watson Ladd, Akamai)

  • Raises question of "backflow" into Linux tree.
  • Answer today: Do not accept PRs on GitHub.

• Dave Thaler: Perhaps the chairs should propose a workflow for issue
  tracking and we just use that.

ISA Extension Policy

https://datatracker.ietf.org/doc/slides-117-bpf-isa-extension-policy/

• Proposal: RFC per extension (set of additions)

  • Allow referencing non-RFC in the meantime to get codepoints?

• Where should registry be: IANA or files in Linux kernel tree?

  • Lars Eggert: if we put it in Linux kernel tree

    • Normative reference: pointing to URL would require special
      process
    • Early allocation of IANA codepoints

  • David Vernet: Can codepoints be expressed in a tree in IANA?

    • Not just 8-bit points, but points that have subpoints

• URI Schemes precedent

  • Tom Herbert: Is it opcodes or more?

    • Dave Thaler: There are a range, more than opcodes. Later
      slide has details.

  • Michael Richardson: Need for range between permanent and
    provisional?

  • Toerless Eckert: Good to have a process that avoids needing to
    change codepoints to change category (provisional -> permanent
    for instance)

ISA registration policies

• Three policies

  • Historical: deprecated things
  • Permanent: standards action
  • Provisional: specificiation required

• Christopher Inacio: problem with provisional policy—when WG closes

• Are the IANA rules we are setting too hard?

  • If we aren't running out of space we can do expert review

• Toerless Eckert: RFC required would provide a fallback

  • Discussion of ISE review impact
  • Proposal withdrawn

• Christoph Hellwig: BPF perspective on process

  • Random people don't submit random extensions
  • Don't block kernel community
  • Standards action seems like a useful default

Hierarchy expression for opcodes in documents

• Option 1: Flat table

• Option 2: Multiple tables

  • Easier to generate option (1) from this format

• Multiple preferences expressed for Option (2)

ABI document bootstrapping

https://datatracker.ietf.org/doc/slides-117-bpf-abi/

Comments

• Christoph Hellwig: ABI standardization will be about minimizing
  differences between platforms.
• David Vernet: Focus is on ISA for now, but no reason not to start
  this.
• Dave Thaler: Useful to have a place to move content not appropriate
  for ISA document.

Collaboration

• Will Hawkins plans to tidy it up and is happy to submit into Linux
  tree

eBPF ELF Profile Specification

https://datatracker.ietf.org/doc/slides-117-bpf-elf-profile-specification/

Preamble

• Intended as informational document to recommend cross-platform
  mechanism.
• Not changing ELF. Using a new "machine" codepoint.

Relationship with BTF

• Note that BTF is proposed standard in working group

• Christoph Hellwig: BTF should be split out, avoid mixing formats of
  different maturity.

  • In favor of different documents for different purposes. Eg. BTF
    can be used without ELF, and also ELF can be used without BTF
    (albeit not recommended)

• David Vernet: A lot of overlap with the ABI document as well.

Conventions

• Various examples for TEXT, DATA sections out there already. Document
  what exists.

• Joe Stringer: How does this intersect with ongoing efforts in
  libraries to converge on approaches?

  • Dave Thaler: This document intends to just gather the state
    right now. Hopefully over time there is less diversity in
    conventions as that will simplify cross-platform
    implementations.

• Let's continue working on this on the list.