Skip to content

run spinxsk on multiple interfaces - #965

Draft
Michael Friesen (mtfriesen) wants to merge 17 commits into
mainfrom
mtfriesen/spinxsk_multi_if
Draft

Michael Friesen (mtfriesen) wants to merge 17 commits into
mainfrom
mtfriesen/spinxsk_multi_if

Conversation

@mtfriesen

@mtfriesen Michael Friesen (mtfriesen) commented Jan 30, 2026

Copy link
Copy Markdown
Contributor

Description

Describe the purpose of and changes within this Pull Request.

To try to increase our repro rate of interesting spinxsk bugs, increase the parallelism of our tests by spinning on multiple NICs. This requires adding support for multiple XDPMP instances, and also adds some incidental improvements to driver cleanup logic after the machine is in a dirty state post-reboot.

If this shows promise, we can try refactoring FNMP to support the same thing.

Testing

Do any existing tests cover this change? Are new tests needed?

Yes.

Documentation

Is there any documentation impact for this change?

No.

Installation

Is there any installer impact for this change?

No.

@mtfriesen Michael Friesen (mtfriesen) added the Area: Tests Related to tests label Jan 30, 2026
Comment thread .github/workflows/ci.yml
@mtfriesen
Michael Friesen (mtfriesen) marked this pull request as ready for review April 15, 2026 11:54
@mtfriesen
Michael Friesen (mtfriesen) requested a review from a team as a code owner April 15, 2026 11:54
Comment thread src/xdp.props
@mtfriesen
Michael Friesen (mtfriesen) marked this pull request as draft April 15, 2026 20:36
Michael Friesen (mtfriesen) added a commit that referenced this pull request Jul 22, 2026
…setup

The socket-setup success-rate breakdown shows setups collapsing at the final bind/activate steps (Prerelease x64: bind 19%, activate 8%; WS2022: bind 30%, activate 20%) while init/umem/rx/tx are healthy. A setup counts as successful only if the socket is fully configured within a 500ms window (completeEvent in QueueWorkerFn). FuzzSocketMisc could issue a blocking XskNotifySocket with a WAIT flag and a timeout of up to 1000ms on the same fuzzer thread doing the configuration, consuming the entire window and preventing bind/activate from completing. Cap the fuzzed notify timeout to <3ms (matching PR #965's mitigation) so setup can progress.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Area: Tests Related to tests

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant