ksummit.lists.linux.dev archive mirror
 help / color / mirror / Atom feed
* kernel.org tooling update
@ 2025-12-10  4:48 Konstantin Ryabitsev
  2025-12-10  8:11 ` Mauro Carvalho Chehab
                   ` (2 more replies)
  0 siblings, 3 replies; 11+ messages in thread
From: Konstantin Ryabitsev @ 2025-12-10  4:48 UTC (permalink / raw)
  To: users, ksummit

Hi, all:

These are the topics that were touched on at the maintainer summit when
discussing tooling on the kernel.org side of things.

--

# What is the state of tooling?

## b4 development update

Past year:

- No major new features in b4 over the past year, excepting `b4 dig`
- Seeing lots of adoption and use across subsystem, with a lot of maintainers
  recommending b4 as the preferred mechanism to submit patches
- Starting to see adoption by several non-kernel projects (openembedded, u-boot, others)
- Significant behind-the-scenes move of codebase to stricter typed code
- Continued work on `b4 review` that got shelved temporarily for other
  priorities.

### LETS PUT MOAR AI INTO IT!!1

I spent a lot of time on trying to integrate AI into b4 workflows, but with
little to show for it in the end due to lackluster results.

- Used local ollama as opposed to proprietary services, with the goal to avoid
  introducing hard dependencies on third-party commercial tooling. This is
  probably the main reason why my results were not so exciting as what others
  see with much more powerful models.

- Focused on thread/series summarization features as opposed to code analysis:

    - Summarize follow-ups (trailers, acks/nacks received), though this is
      already fairly well-handed with non-AI tooling.

    - Gauge "temperature" of the discussion to highlight controversial series.

    - Gauge quality of the submission; help decide "is this series worth
      looking at" before maintainers spend their effort looking at it, using
      maintainer-tailored prompts. This may be better done via CI/patchwork
      integration, than with b4.

    - Use LLM to prepare a merge commit message using the cover letter and
      summarizing the patches.

I did not end up releasing any features based on that work, because:

    - LLM was not fantastic at following discussions and keeping a clear
      picture of who said what, which is kind of crucial for maintainer
      decision making.

    - Very large series and huge threads run out fo context window, which
      causes the LLM to get even worse at "who said what" (and it's
      already not that great at it).

    - Thread analysis requires lots of VRAM and a modern graphics card, and is
      still fairly slow there (I used a fairly powerful GeForce RTX).

    - Actual code review is best if it happens post-apply in a temporary
      workdir or a temporary branch, so the agent can see the change in the
      context of the git tree and the entire codebase, not just the context
      lines of the patch itself.

I did have much better success when I worked to represent a thread not as
multiple messages, but as a single document with all interleaved follow-up
conversations collated together. However, this was done manually --
representing emails from arbitrary threads as such collated documents is a
separate challenge.

Using proprietary models and remote services will probably show better
results, but I did not have the funds or the inkling to do it (plus see the
concern for third-party commercial tooling). I may need to collaborate more
closely with the maintainers already doing it on their own instead of
continuing my separate work on it.

### AI crawler scourge

While working on LLM integration, it was certainly ironic that one of the
top challenges for us was to try to keep AI crawlers from overwhelming
kernel.org infrastructure. While we've put several mitigations in place, it's
a temporary relief at best.

## Continuous degradation of SMTP

We're increasingly having to deal with the degradation of the SMTP support by
all commercial companies:

    - major hosts are increasingly not interested in getting mail from anyone
      who isn't also a major mail service provider

    - their "bulk sender" guidelines are no good for us (e.g. requiring that
      we add one-click unsubscribe footers to all email)

    - their "spam filters" are increasingly based on training data, which
      means that "looks different from what most of our users receive" is
      enough to have patches and code discussions put into the "Junk" folder

    - they apply arbitrary throttling ("too many deliveries for the same
      message-id", "too many messages from the DKIM domain foobar.com")

    - anti-phishing services at commercial IT companies do horrible things to
      incoming messages

## Are we finally moving away from patches sent over email?

There are still important concerns when we consider moving away from "patches
sent via email":

    - SMTP is still the only widely used protocol we have for decentralized
      communication; everything else is experimental or has important
      drawbacks, such as:

        - it relies on single-point-of-failure services (e.g. Signal), or
        - it requires standing up esoteric software (which then become
          single-point-of-failure services), or
        - it requires an "everyone-must-switch-now" flag day

    - RFC-5322, with all its warts, is a well-defined standard for internet
      messages:

        - robust, capable of dealing with change while preserving legacy
        - easy to parse with libraries for almost any framework
        - easy to archive and query
        - has lots of tooling built around it

With lore and public-inbox, we *are* in the process of moving away from
relying on the increasingly unreliable SMTP layer. Lore can already let you do
the following things:

    - lets anyone submit patches via the web endpoint
    - lets anyone subscribe to lists via several protocols (NNTP, POP, IMAP)
    - lets anyone use lei to receive arbitrary feeds
    - can aggregate any number of sources, as long as they are RFC-5322
      messages (or can be converted to them)

Lore and public-inbox is becoming a kind of a distributed, replicating
messaging bus with a robust query and retrieval interface on top of it, and I
believe it's a fairly powerful framework we can build upon.

## Work on "local lore"

One downside of lore.kernel.org is that it's a central service, which runs
counter to our goal of limiting how many single points of failure we have.
There is fault-tolerance built into the system (lore.kernel.org is actually 4
different nodes in various parts of the world), but an adversary would have no
difficulty knocking out all nodes at once, which would impact the project
significantly.

The "local lore" projects it the attempt to provide a kind of "maintainer
container" that can be run locally or in any public cloud:

    - comes with a 6-month constantly-updating mirror of lore, using a
      failover set of replication URLs (including tor/onion)
    - comes with a pre-configured mirror of git repositories that are kept
      up-to-date in the same fashion
    - lets the maintainer set up lei queries that can push into their
      inbox, supporting Gmail+OAuth, JMAP, IMAP
    - provides a web submission endpoint and an SMTP service that can
      integrate with other SMTP relays
    - publishes a public-inbox feed of maintainer activity that central
      lore can pick up and integrate

There is a parallel goal here, which is to make it easier for devs to assume
maintainer duties without having to spend a week setting up their tooling.
In theory, all they would be need to do is to set up their maintainer
container and then use the web menu to choose which feeds they want to pull
and where they want messages delivered.

This project is still early in development, but I hope to be able to provide
test containers soon that people can set up and run.

## Other tools

### Bugzilla

It may be time to kill bugzilla:

    - despite periodic "we're not dead yet" emails, it doesn't appear very
      active
    - the upgrade path to 6.0 is broken for us due to bugzilla abandoning the
      5.2 development branch and continuing with 5.1
    - question remains with what to replace bugzilla, but it's a longer
      discussion topic that I don't want to raise here; it may be a job for
      the bugspray bot that can extend the two-way bridge functionality to
      multiple bug tracker frameworks

### Patchwork

Patchwork continues to be used widely:

    - we've introduced query-based patchworks, where instead of consuming the
      entire mailing list, we feed it the results of lei queries
    - I'm hoping to work with upstream to add a couple of features that would
      be of benefit to us, such as:

        - support for annotating patches and series (e.g. with LLM summaries)
        - an API endpoint to submit patches, so maintainers could add
          arbitrary series to their patchwork project, integrating with b4

## Web of Trust work

There is an ongoing work to replace our home-grown web of trust solution (that
does work but has important bottlenecks and scaling limitations) with
something both more distributed and easier to maintain. We're working with
OpenSSF to design the framework and I hope to present it to the community in
the next few months.

## Questions?

Send away!

-K

^ permalink raw reply	[flat|nested] 11+ messages in thread

end of thread, other threads:[~2025-12-18 14:09 UTC | newest]

Thread overview: 11+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2025-12-10  4:48 kernel.org tooling update Konstantin Ryabitsev
2025-12-10  8:11 ` Mauro Carvalho Chehab
2025-12-10 13:30 ` Thorsten Leemhuis
2025-12-11  3:04   ` Theodore Tso
2025-12-12 23:48   ` Stephen Hemminger
2025-12-12 23:54     ` Randy Dunlap
2025-12-16 16:21 ` Lukas Wunner
2025-12-16 20:33   ` Jeff Johnson
2025-12-17  0:47     ` Mario Limonciello
2025-12-18 13:37       ` Jani Nikula
2025-12-18 14:09         ` Mario Limonciello

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox